Investigating the various impacts of COVID-19 using Sentiment Analysis and Topic Modeling over three years

Investigating the various impacts of COVID-19 using Sentiment Analysis and Topic Modeling over three years | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Investigating the various impacts of COVID-19 using Sentiment Analysis and Topic Modeling over three years Mustapha Hankar, Mohammed Kasri, Toufik Mzili, Abderrahim Beni-Hssane This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5435843/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The world health organization (WHO) has officially declared the COVID-19 as a global pandemic in March 2020. Consequently, nations worldwide took some preventive measures, including lockdowns, quarantines, and social distancing to slow down the spread of coronavirus. This unprecedented event has profoundly disrupted the normal way of life. The pandemic had devastating impacts on various aspects of society such as healthcare systems, social life, the economy, and education. People from around the world began expressing emotions of fear, isolation, and various kinds of traumatic disorders on social media networks such as Twitter and Facebook. This research paper explores the impacts of COVID-19 in Morocco using topic modeling, sentiment analysis, and time series analysis. The study follows a two-step process. Initially, we employed a topic model, specifically BERTopic, to extract the main themes from a dataset containing comments gathered from the online newspaper Hespress and Twitter. Subsequently, we conducted a topic-based sentiment analysis to assess how COVID-19 has impacted Moroccans through a time window of three years. The findings revealed that sentiments related to the various topics were highly negative. In addition, we leveraged time-series data on COVID-19 to examine how the evolving epidemiological situation influenced sentiments from March 2020, the beginning of the pandemic, until the end of 2022. Our analysis indicated a strong correlation between changes in COVID-19 cases and sentiment analysis results. Topic modeling Time series Sentiment analysis COVID-19 impacts Twitter Hespress Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 1. Introduction Please have a look at courier new font provided for text in article. The new coronavirus, also called SARS-CoV-2, has created a calamitous situation around the globe by causing millions of deaths, and infections and putting other lives at stake. The pandemic has severely affected and challenged public healthcare systems, governments, and societies in an unimaginable way. Moreover, its impacts still go on and will remain for many years to come[ 1 , 2 ]. COVID-19 outbreak had a negative impact on every aspect of life including public health, economy, education, travel, social life, labor market, etc.[ 3 – 6 ]. As the pandemic continues to spread exponentially and new variants emerge, it is clear that the world will be facing a long-term crisis and disruption mostly on the economic level. In the periods of quarantine and stay-in-home measures, people have suffered from mental health disorders such as stress, anxiety, depression, feelings of isolation, worry, etc. Many conducted studies[ 7 – 9 ] showed that protective measures which have been imposed by governments such as travel restrictions, lockdowns, and economic shutdowns have exposed people to psychological disorders while being isolated in their places, socially distanced from one another, and living at a constant risk of losing their work. These effects include symptoms of post-traumatic stress disorder, long quarantine suppress symptoms, and many others. Social media outlets like Facebook and Twitter, news platforms, blogs, and forums, during the periods of the pandemic have been experiencing reactions, feelings, emotions, and thoughts of people towards the COVID-19 pandemic and its impacts [ 10 ]. Natural language processing (NLP) techniques are broadly applied to retrieve and analyze relevant information from these huge amounts of generated content. For instance, topic models are used in this context to extract the main topics related to the outbreak. Sentiment analysis is another NLP technique that is often applied to identify sentiments and emotions in a given text as a way to investigate the effects of COVID-19 on people’s normal lives. Sentiment analysis is widely applied across various domains like business intelligence to analyze customers feedback regarding a service or a product [ 11 ], and social media monitoring to analyze people’s opinions about public policies or certain figures[ 12 , 13 ]. Sentiment analysis approaches are mostly categorized into three main levels: document level, sentence level, and aspect level [ 14 ]. In document level, the whole document is classified as whether negative, neutral, or positive. In sentence level, subjective sentences are filtered out before classifying their polarity. The aspect-level approach focuses on performing sentiment analysis regarding a specific aspect [ 15 ]. In this paper, we performed a topic-based sentiment analysis on a corpus of collected comments related to COVID-19 outbreak. The dataset that is made available for this study consists of comments scraped from the online newspaper Hespress † written in both modern standard Arabic (MSA) and Moroccan dialect (MD). Our approach follows a process of two steps: topic extraction and sentiment analysis. For topic modeling, we applied an embedding-based topic model, namely BERTopic, to retrieve topics from the corpus and generated document clusters based on the extracted themes. Second, a topic-based sentiment analysis is carried out to identify polarities within the texts and classify them into positive, neutral, or negative. Finally, a time series analysis is performed on a monthly basis to track changes in sentiments in the period of three years of the pandemic. The visualization of these changes shows how people's reactions and sentiments evolve over time regarding the COVID-19 situation. Time series sentiment analysis can relate events like quarantine, COVID-19 spikes of infections and deaths, travel restrictions, etc.; to the changes in sentiments and emotions of people. For example, our study showed that negative sentiments of fear, anxiety, and depression are increased during quarantine periods and when a new wave of COVID-19 variant emerges and spreads. Negative sentiments also increased regarding the vaccine topic indicating that people hold suspicious sentiments towards vaccination campaigns. The key contributions to this paper are: (i) collecting a real dataset containing thousands of comments written in Arabic, (ii) designing a semi-supervised model that integrates both topic modelling and sentiments analysis techniques to examine the effects of COVID-19 in Morocco, and (iii) tracking the changes of sentiment analysis results from March 2020 until the end of December 2022 in correlation to the epidemiological situation. The rest of the article is organized as follows: The second section is dedicated to review and discuss related works to the study. In section 3, we introduce the methods and the materials used to perform this research. Results are presented and discussed in section 4. Finally, we summarize our paper and discuss some feature works in the section 5. [ † ] Hesspress news website (accessed 26/5/2024): https://www.hespress.com/ 2. Related Works In recent years, academic researchers have shown growing interest in the field of NLP in order to design advanced techniques for processing human languages. However, Arabic NLP is still facing many challenges as a result of some specific and intrinsic features of the language. This section will present and examine some relevant research that relates to our study. Chandrasekaran et al.[ 16 ] analyzed trends, topics, and sentiments in a dataset that contains tweets about COVID-19 outbreak. They first applied Latent Dirichlet Allocation (LDA) to extract topics within the dataset and then implemented a dictionary-based method, namely valence aware dictionary and sentiment reasoner (VADER), to compute sentiment scores. In another work, Xue et al. [ 17 ] analyzed COVID-19 tweets using LDA and machine learning-based sentiment analysis to examine public discourse and reactions of tweeters to the pandemic. Their results showed that people expressed feelings of fear and threat regarding the unknown nature of the coronavirus virus. Boon-Itt and Skunkan [ 18 ] conducted a study by utilizing LDA and sentiment analysis techniques to explore and identify topic discussions in a dataset of collected Tweets over time. In their experiment results, they claimed that people expressed negative feelings towards the COVID-19 outbreak. Yin et al. performed [ 19 ] an in-depth analysis of social media posts and discussions related to COVID-19 vaccines on Twitter. In the first step, the authors used LDA to extract the main topics about various vaccines and then applied VADER model to compute sentiment polarities. Reported results showed that people feel positively confident to take vaccines. Qorib et al [ 20 ] conducted a study to examine COVID-19 vaccine hesitancy among people through their social media posts. They applied several machine learning algorithms combined with different vectorization methods (TF-IDF, Doc2vec, and BoW) to predict hesitancy regarding vaccines. In their experiment results, they suggest that the combination of TextBlob, TF-IDF, and LinearSVC outperformed other models in sentiment classification. They also concluded that people are feeling optimistic about taking vaccines and hesitancy decreases over time. Abdul-Mageed et al. [ 21 ], developed a hybrid model for subjectivity detection and sentiment analysis. The model first detects subjectivity within a text and then classifies its polarity into neutral, positive, or negative. They tested the model on a dataset collected from social media outlets like Twitter and other web platforms such as Wikipedia Talk Pages, mini blogs, chat apps, and web forums. The model utilizes a machine learning algorithm (SVMlight) to classify sentiments. However, they faced challenging and complex issues in their experiments due to the specific characteristics of the Arabic language. In [ 22 ], Zarra et al. introduced a semi-supervised model that integrates topic modeling with sentiment analysis to conduct aspect-oriented opinion mining. They used an unsupervised method to retrieve the main themes within Facebook comments written in colloquial Maghreb and then applied a supervised technique to compute sentiment polarity. Shelke et al.[ 23 ] designed an aspect-oriented model to conduct sentiment analysis on a dataset of product reviews. They used SentiWordNet lexicon and some specific features of reviews to identify their polarity. Madani et al.[ 24 ] designed a sentiment analysis recommender to classify tweet as positive, negative, or neutral. The dataset they used for the study contains COVID-19 related tweets from Morocco, collected between March 2020 and October 2020. Their experimental results showed that their proposed model reached 86% accuracy outperforming baseline machine algorithms. They found out that changes in sentiments over time are affected by the COVID-19 epidemic situation. In our previous works [ 25 , 26 ], we used LDA to extract topics from a collected dataset that contains more than 20,000 comments from Hespress, and then we applied a pretrained transformer model to classify sentiments into negative, positive, and neutral. The findings showed that negative sentiments were high regarding all extracted topics in the first year of the pandemic. 3. Proposed Approach 3.1 Datasets 3.1.1 COVID-19 Time Series Many sources such as the Moroccan ministry of health, WHO, and the university of Johns Hopkins contributed to the collection of this dataset. Since January 22, 2020, the team at the at Johns Hopkins Center for Systems Science and Engineering (CSSE) has been responsible for recording and updating COVID-19 data from across the globe. They have been cleansing, transforming, and normalizing data to make it easier for analysis and further processing. They arranged dates and consolidated several files to transform data into normalized time series. The dataset is made available in a GitHub data repository as a CSV file and can be accessed via this link ‡ . The dataset contains six columns: the dates that confirmed cases or fatalities were recorded, cumulative confirmed cases, cumulative deaths, recovered cases, Region/Country, and finally Province/state. We filtered out the time series based on the country column to retrieve cases and fatalities based in Morocco from March 2, 2020, the date in which the first case appeared, to December 31, 2022. We transformed the dataset into a time series of daily confirmed cases and daily fatalities indexed by DateTime. 3.1.2 Hespress Comments The dataset contains more than 20,000 comments sourced from the prominent news outlet, Hespress. These comments are written in both Modern Standard Arabic (MSA) and Moroccan Dialect (MD). For data collection, we developed a Python crawler using the Selenium library to extract comments from news articles related to the COVID-19 pandemic across various domains, including health, economy, politics, society, and vaccines. The scraping process was conducted during different time frames between March 2020 and December 2022 to ensure comprehensive coverage of the COVID-19 timeline. Comments in languages other than Arabic, such as English, French, or Spanish, were filtered out to retain only Arabic comments. The resulting dataset is stored as a CSV file, containing columns for the publication date, the user's name, and the comment text. The dataset can be accessed on GitHub via this link § . Arabic language processing is still facing numerous challenges due to some inherit characteristics of the language itself like metaphor, diglossia, ambiguity, and various other difficulties related to Arabic morphology [ 27 ]. Moreover, researchers are frequently encountered with a mixture of three versions of the Arabic script on the web: Modern Standard Arabic, Classical Arabic, and Dialectical Arabic. The differentiations in these scripts make processing and understanding tasks more difficult. Arabic language has a different structure and morphology compared to Latin-based languages. For instance, Arabic script is written from right to left and has 28 different letters. The Arabic letters can have diacritics such as hamza(همزة:ء), sukun(سكون: ْ), fatha(فتحة: َ), kasra(كسرة: ِ), tanwin(تنوين: ً), etc. These diacritics can directly affect the semantics of words in the text and increases ambiguity to capture its meaning. The morphology of Arabic letters can change according to their location (initial, medial, final, or isolated) in the text which makes the processing task even more challenging. Cleaning texts in Arabic includes text normalization and removing punctuations, diacritics, numbers, links, and elongations [ 28 ]. The preprocessing task starts with tokenization to split the text into tokens, stop words removal to eliminate non-informative words such as fi (في), min (من), ala (على), etc., and stemming to transform words into their base form by removing suffixes and prefixes [ 27 ]. Table 1 below shows a sample of texts in the dataset before and after preprocessing and Fig. 1 illustrates the main steps to preprocess Arabic text. Table 1. Examples of comments before and after preprocessing. Raw text Preprocessed text كلنا متضررين و لكن أحسن حاجة هي تمديد الحجر الصحي مع مراعاة الظروف الإجتماعية لينا كاملين. و الله يحد الباس. متضررين احسن حاجة تمديد حجر صحي مراعاة ظروف اجتماعية كاملين الله يحد باس الدار البيضاء هي اكثر المدن كثافة، هي بمثابة مغرب مصغر وهي قلب الاقتصاد الوطني .اظن انه يجب تخفيف الحجر على باقي الجهات التي لم تعد تسجل اصابات عديدة والتركيز على البيضاء و خاصة الشركات اما انتظار تسجيل 0 اصابة للرفع فهو ضرب من. ضروب المستحيل الدار البيضاء مدن كثافة بمثابة مغرب مصغر قلب اقتصاد وطني اظن يجب تخفيف حجر باقي جهات لم تعد تسجل اصابات عديدة تركيز شركات انتظار تسجيل اصابة رفع ضرب ضروب مستحيل بالنسبة للناس اللي ما خداوش جرعة اللقاح راه ما عندهمش اي نية للسفر سواء داخل المغرب او خارجه فاغلب الناس المهتمين بالسفر تلقحو هادشي راه ماشي معقول لا حول ولا قوة إلا بالله ناس ماخداوش جرعة لقاح ما عندهمش نية سفر سواء داخل المغرب خارج اغلب ناس مهتمين سفر تلقحو ماشي معقول Arabic text may include Arabic numerals, Arabic-specific characters, or mixed Arabic words with words in languages. Handling these challenges involves making decisions based on the specific requirements of the sentiment analysis task. For example, we might choose to replace Arabic numerals with their written forms or decide how to handle mixed language text and code-switching. Arabic text may contain spelling errors or typos. Applying spell checking and correction techniques can improve the accuracy of subsequent analysis steps. Arabic-specific spell checkers or general-purpose spell checkers with Arabic language support can be used. Additional preprocessing steps may be required for the Arabic language. For example, for aspect-based sentiment analysis, we may try identifying and extracting aspect terms or entities from the text using named entity recognition or part-of-speech tagging. 3.1.3 TBCOV Tweets TBCOV, also known as two billion COVID-19, is a large-size Twitter collection [ 29 ] that contains more than 2B multilingual COVID-19-related tweets. Specifically, TBCOV contains over two billion tweets using more than 800 keywords from different languages. The collection was gathered within a period of 14 months beginning from February 1, 2020, until March 31, 2021. The tweets are written in 67 different languages (including Arabic) and posted by more than 87 million unique users on Twitter across 218 countries all over the world. Due to the largeness of the original multilingual dataset, the collectors have offered different filters based on location, language, and time window on their website ** . We used language and location filters to retrieve more than 40,000 Arabic-written tweets that were published in Morocco during the same period of collection. The Arabic tweets are cleaned, preprocessed, and prepared for topic modeling and sentiment analysis using NLP tools. 3.2 Topic Extraction using BERTopic Topic modeling is a text-mining method that is commonly used to discover the hidden themes within a corpus of textual documents. Topic models represent documents in the collection as a mixture of different topic, and in turn, each topic is considered as a distribution of various words weighted by their respective scores. Topic models are widely applied in several domains like information retrieval, text classification, sentiment analysis, and recommendation systems, etc. Classic topic models like LDA[ 30 ] and non-negative matrix factorization (NMF)[ 31 ] are limited to representing documents as a bag of words (BoW) which leads to ignoring the order of words, semantic relationships, and contextuality between words [ 32 , 33 ]. In response to this issue, word embedding models have emerged in NLP field to address the problems of BoW [ 34 , 35 ]. For instance, Bidirectional Encoder Representations from Transformers (BERT)[ 36 ] have shown promising results in generating contextual text representations that capture the semantic structures within sentences and word vectors. Modern topic models take great advantage of the of word embeddings to build coherent models powered with centroid-based techniques. For example, BERTopic generates representations of topics in documents in a way that each topic is assigned to a cluster of documents. The term frequency-inverted document frequency (TF-IDF) is computed by multiplying term frequency \(\:t{f}_{t,d}\) of a word t in document d by the inverse document frequency (IDF). The value of IDF is derived by calculating the logarithm of the corpus size N divided by the total number of documents containing the term t. The formula is shown in Eq. ( 1 ) below: $$\:{\text{t}\text{f}\text{i}\text{d}\text{f}}_{t,d}={\text{t}\text{f}}_{t,d}\times\:\text{l}\text{o}\text{g}\left(\frac{\text{N}}{{\text{d}\text{f}}_{t}}\right)$$ 1 This TF-IDF formula is adjusted to measure the importance of a term to a topic instead of a document. The Eq. ( 2 ) measures the class-based TF-IDF of a term in a given class: $$\:{W}_{t,c}=t{f}_{t,c}\cdot\:\text{l}\text{o}\text{g}\left(1+\frac{A}{t{f}_{t}}\right)$$ 2 where \(\:t{f}_{t,c}\) measures the frequency of a term t in class c representing a set of documents grouped into one document known as a cluster. This term frequency is then multiplied by the inverse class frequency (ICF) to assess the importance of a term within a class. The value of ICF is obtained by computing the logarithm of the average number of terms in each class A divided by the frequency of the term t across classes. The one inside the logarithm is added to the division to ensure the output score values remain positive. The class-based TF-IDF assesses the relevance of terms in document clusters allowing to extract topic-term distributions for each cluster. In this paper, we applied a word embedding-based technique, namely BERTopic, to extract topics from Hespress comments and tweets to generate document clusters. Afterward, each document cluster is assigned to each extracted topic based on the aspect representations. As for that, we used BERTopic due to its good results in generating coherent and diverse topics when compared to the performance of conventional topic models such as LDA[ 30 ], NMF[ 31 ], Doc2vec[ 37 ], and Top2vec[ 38 ] in various Arabic datasets[ 39 ] including Hespress comments dataset. Table 2 shows experimental results of four topic models in terms of topic coherence (TC) and topic diversity (TD). TC measures the rate of semantic similarity between the most important words in a topic while TD measures the degree of diversity among topics. Table 2 Performance of topic models leveraged with topic coherence and topic diversity across different topic numbers (K). Topic Model K = 6 K = 10 K = 15 TC TD TC TD TC TD LDA -0.021 0.42 -0.029 0.28 -0.024 0.33 NMF 0.158 0.95 0.18 0.93 0.196 0.92 Topic2Vec 0.093 0.84 0.099 0.72 0.154 0.672 BERTopic 0.22 0.96 0.211 0.992 0.20 0.99 From Table 2, we can see that BERTopic outperforms other topic models with high scores of topic coherence and topic diversity. The results shown in the table were recorded after five iterations of training across three different topic numbers (K=6, K=10, K=15) to avoid overfitting. The word embeddings of texts were generated using an Arabic pre-trained word embedding model, namely AraBERT [40]. Afterward, we trained the BERTtopic model on top of the constructed word embeddings to extract the themes that are related to the COVID-19 pandemic. The extraction of topics is performed in the Algorithm 1. 3.3 Sentiment Classification using CAMeL Sentiment analysis (SA) is an NLP technique which is often used to identify sentiments within subjective text and classify its polarity into negative, neutral, or positive. Sentiment analysis techniques are typically categorized into three primary types: lexicon-based techniques, machine learning techniques, and hybrid approaches[ 15 ]. Machine learning techniques combine statistical and probabilistic algorithms with specific linguistic characteristics to identify sentiments in a given text. However, Arabic sentiment analysis (ASA) is particularly facing numerous challenges because of various language-specific characteristics. To address these challenges, researchers have contributed to developing various tools and resources, such as CAMeL [ 41 ]. This open-source toolkit is used to perform various Arabic NLP tasks including sentiment analysis, part-of-speech tagging, and name entity recognition. CAMeL sentiment analyzer is a pre-trained model that is fine-tuned on AraBERT and multilingual BERT word embeddings to detect sentiments in Arabic texts. The classifier was trained and evaluated on different datasets including the Arabic Speech-Act dialectical dataset and the ArSAS tweet sentiment corpus. Table 3 shows the evaluation results of the CAMeL sentiment analyzer over three datasets. Table 3 CAMeL sentiment analyzer accuracy using AraBERT and mBERT compared to Mazajak over three benchmark datasets [ 41 ]. CAMeL(AraBERT) CAMeL(mBERT) Mazajak ArSAS 0.92 0.89 0.90 ASTD 0.73 0.66 0.72 SemEval 0.69 0.60 0.63 3.4 Topic-based Sentiment Analysis The conventional models employed for Arabic sentiment analysis often focus on computing text polarity on the document level while ignoring the aspect/topic level that holds an opinion towards a specific entity within the text. In our approach, we consider that subjective texts express multiple topics and each one of them has its own sentiment. Given this, we chose to perform sentiment analysis on a topic-based level to capture people's opinions and examine their reactions regarding many different COVID-19 related aspects or topics. Our approach follows a process of two steps: we first extract topics from the corpus of documents using BERTopic and then generate word distributions of extracted topics based on word scores representing each specific topic. Afterward, document clusters are constructed and assigned to map each extracted topic with a corpus of documents using the c-TF-IDF formula. Table 4 presents the final topics and their word distributions while Fig. 2 shows the distribution of document clusters for each topic. Table 4 High-scored words for each topic. Topic High-scored topic words Economy محلات | متاجر | شغل | صناعة | معامل | اقتصاد | عمل | اغلاق | شركات | أرباب Education دراسة | تعليم | عمومي | عن بعد |حجر صحي | استاذ | امتحان | مدرسة | منصة | تربية | وزارة | تلميذ Fear جوع | ملل | خطر | موت | بلاء | قلق | خوف | عدوى | شفاء | الله | سقم | انتحار| مصيبة | تشاؤم | صحة Support تمديد | تعويضات | راميد | صندوق | دعم | ازمة | بطالة | ضمان | مساعدات | حجر صحي | داخلية | طوارئ | التزام | دولة | حكومة | مواطن | حظر Health تعقيم | اعراض | فقدان | شم | ذوق | تنفس | نقص | اختناق | وقاية | حالات | صحة | تحاليل | اصابات | كمامة Vaccine جواز | دلتا|لقاح | كوفيد | اوميكرون | سينوفارم | استرازينيكا | فايزر | فيروس | تلقيح | متحور | الغاء | جرعة خطة | مسرحية | كذب | اوهام | مؤامرة | ماسونية | ابادة | تأثير In the second step of our approach, we transformed constructed document clusters and saved them as data frames and each data frame represents a document cluster of a specific topic. We mention that a text in the corpus can belong to multiple clusters at once and thus it may hold many opinions about different topics. After that, we used CAMeL to classify texts of document clusters into positive, neutral, or negative. The results of the whole process from text representation to topic clustering to sentiment analysis is a refined aspect-based infodemic sentiment analysis. Algorithm 2 represents the performed sentiment classification task. Finally, we merged the data frames with the COVID-19 time series to track sentiment changes over two periods of time and to examine any correlations between the epidemiological situation in Morocco and the variation in sentiment results. Figure 3 shows the overall workflow of our proposed approach. [‡] C0VID-19 time series data repository (accessed 25/5/2024): https://github.com/CSSEGISandData/COVID-19 [§] Hespress dataset repository (accessed 26/5/2024): https://github.com/HankarM88/Hespress_COVID-19_Dataset [**] TBCOV dataset (accessed 26/5/2024): https://crisisnlp.qcri.org/tbcov 4. Results and discussion This section outlines the experiment results of applying our proposed approach. While most approaches in the literature review have examined the question of COVID-19 pandemic effects on people by simply applying baseline sentiment analysis methods to identify sentiments in the user-generated content from Twitter, Facebook, and other social media outlets; we combined many techniques including topic modeling, word embeddings, sentiment analysis, and time series to realize this work. Table 5 below shows the frequency of sentiment polarities and the polarity rate regarding each topic. Table 5 topic-level sentiment analysis results Topic Polarity Frequency Percentage (%) Economy Positive 254 6.3 Neutral 537 13.3 Negative 3285 80.5 Education Positive 480 8.8 Neutral 849 15.6 Negative 4114 75.6 Fear Positive 1550 17.09 Neutral 1118 12.4 Negative 6400 70.6 Health Positive 631 8.85 Neutral 1198 16.8 Negative 5300 74.5 Support Positive 344 7.7 Neutral 524 11.7 Negative 3604 80.6 Vaccine Positive 283 3.8 Neutral 1252 17 Negative 5815 80.2 The overall findings showed that COVID-19 outbreak has dramatically affected most aspects of normal life in Morocco, meaning that people had reacted with high rates of negative sentiments regarding all COVID-19 related topics. For example, more than 80% of respondents were feeling very deficient on how the outbreak has calamitously affected the economy and job market, 70% expressed feelings of anger, fear, and anxiety towards the novel coronavirus spread, and 80% of them expressed negative feelings about education, especially during lockdowns where students had to stay at home and learn online. People have worried about health systems collapse because of the huge number of new cases received every day at hospitals. The majority of Moroccans reacted to vaccine campaigns (80%) and expressed concerns of worry and hesitancy about the ones that the health authorities suggested for people to take, especially Sinopharm and AstraZeneca vaccines. In the first year of the pandemic in Morocco (2020), people have been mostly concerned about health and education. This led them to express emotions of fear and disruption as illustrated in Fig. 4 . However, high rates of negative concerns and reactions have emerged about the vaccine in the second year of the pandemic (Fig. 5 ). The widespread dissemination of misleading information and fake news about COVID-19 outbreak and the vaccines through social media and the internet has led to a growing sense of fear and distrust among people, making them hesitant to take the vaccine. This can be seen clearly in Fig. 6 which shows the impacts of COVID-19 regarding all other topics except vaccine. sentiments of fear were highly negative for these topics. Consequently, some people have formed groups of anti-vaccination to manifest against vaccines, and imposing mandatory vaccine passes for accessing government buildings, and other public places. As a result, vaccine hesitancy has become a pressing issue that drives people to delay in accepting or refusing vaccines despite their availability. It is often fueled by concerns, doubts, and misinformation shared on social media and the internet. This hesitancy can lead to lower vaccination rates, hindering efforts to achieve herd immunity and control the spread of the virus. We intended to separate our study into three different periods of the pandemic to track sentiment changes and compare these variations from March 2020 until the end of 2022 in the history of the pandemic. Figure 7 shows that negative sentiment rates started to increase exponentially from March 2020, the very beginning of the outbreak, to peak in November 2020. After this period, people’s general concerns about the COVID-19 situation started to fade. In the second period, negative sentiment feedback was not as high as in 2020. However, it is obvious from Fig. 8 that negative sentiments in the time window of March 2021 to May 2021 have spiked and after that started to decrease. After June 2021, negative polarities started to increase slightly until November and decreased at the end of the year. In the beginning of 2022, negative sentiments continued to decrease until March and then started to increase and peaked in June and started to decrease until the end of the year while rates of positive sentiments slightly increased in the last months of the year. The variation of negative sentiments over time is well explained in Figs. 10, 11 , and 12. The epidemiological situation, represented by monthly records of COVID-19 cases and fatalities, directly influences how people react to the pandemic. The line plots show that changes in COVID-19 situations directly affect feedback sentiments during a given period of time. High rates of negative sentiments are explained by the spikes of coronavirus spread (cases and fatalities) among people and the subsequent actions taken, such as lockdowns, wearing masks, social distancing, and other protective measures to slow down the outbreak. As the infection rate decreases and the government slightly eases protective measures, negative sentiments also decrease, and people gradually return to their normal lives. When the situation improves, and there is a lower risk of COVID-19 transmission, the negative sentiments may start to subside. People might feel more relieved and less anxious as they perceive a lower threat level. As a result, they may gradually return to their pre-pandemic routines and lifestyles. The interplay between the epidemiological situation, government actions, and people's reactions can significantly impact the changes in sentiments overtime during the course of the COVID-19 pandemic. 5. Conclusion In this paper, we proposed an approach that combines many different NLP techniques to examine COVID-19's effects on Moroccan citizens through their generated content in social media and other web platforms. The system takes textual data from different sources (Twitter and the online newspaper Hespress) and performs necessary text cleaning and preprocessing, then extracts main topics within the corpus using the BERTopic model, and finally identifies sentiments on a topic-based level using the pre-trained Arabic CAMeL analyzer. The results showed that the pandemic has drastically affected all aspects of normal life. We also used COVID-19 times series data of monthly cases and fatalities to examine sentiment variation within three different periods in the outbreak’s history, and we found that the epidemiological situation highly impacts sentiment analysis results. References Haleem A, Javaid M, Vaishya R (2020) Effects of COVID-19 pandemic in daily life. Curr Med Res Pract 10. https://doi.org/10.1016/j.cmrp.2020.03.011 Onyeaka H, Anumudu CK, Al-Sharify ZT, Egele-Godswill E, Mbaegbu P (2021) COVID-19 pandemic: A review of the global lockdown and its far-reaching effects. Sci Prog 104. https://doi.org/10.1177/00368504211019854 Škare M, Soriano DR, Porada-Rochoń M (2021) Impact of COVID-19 on the travel and tourism industry. Technol Forecast Soc Change 163. https://doi.org/10.1016/j.techfore.2020.120469 Priya SS, Cuce E, Sudhakar K (2021) A perspective of COVID 19 impact on global economy, energy and environment. Int J Sustain Eng 14. https://doi.org/10.1080/19397038.2021.1964634 Chang AY, Cullen MR, Harrington RA, Barry M (2021) The impact of novel coronavirus COVID-19 on noncommunicable disease patients and health systems: a review. J Intern Med 289. https://doi.org/10.1111/joim.13184 Pokhrel S, Chhetri R (2021) A Literature Review on Impact of COVID-19 Pandemic on Teaching and Learning, Higher Education for the Future. 8. https://doi.org/10.1177/2347631120983481 Passavanti M, Argentieri A, Barbieri DM, Lou B, Wijayaratna K, Foroutan Mirhosseini AS, Wang F, Naseri S, Qamhia I, Tangerås M, Pelliciari M, Ho CH (2021) The psychological impact of COVID-19 and restrictive measures in the world. J Affect Disord. https://doi.org/10.1016/j.jad.2021.01.020 Serafini G, Parmigiani B, Amerio A, Aguglia A, Sher L, Amore M (2020) The psychological impact of COVID-19 on the mental health in the general population. QJM 113. https://doi.org/10.1093/qjmed/hcaa201 Brooks SK, Webster RK, Smith LE, Woodland L, Wessely S, Greenberg N, Rubin GJ (2020) The psychological impact of quarantine and how to reduce it: rapid review of the evidence, The Lancet. https://doi.org/10.1016/S0140-6736(20)30460-8 Nemes L, Kiss A (2021) Social media sentiment analysis based on COVID-19. J Inform Telecommunication 5. https://doi.org/10.1080/24751839.2020.1790793 Quyyam T, Ghous H (2021) Sentiment Analysis of Amazon Customer Product Reviews: A Review. Int J Sci Res Eng Dev. 4 Chakraborty K, Bhattacharyya S, Bag R (2020) A Survey of Sentiment Analysis from Social Media Data. IEEE Trans Comput Soc Syst 7. https://doi.org/10.1109/TCSS.2019.2956957 Kasri M, El-Ansari A, El Fissaoui M, Cherkaoui B, Birjali M, Beni-Hssane A (2023) Public sentiment toward renewable energy in Morocco: opinion mining using a rule-based approach. Soc Netw Anal Min 13:124. https://doi.org/10.1007/s13278-023-01119-3 Birjali M, Kasri M, Beni-Hssane A (2021) A comprehensive survey on sentiment analysis: Approaches, challenges and trends. Knowl Based Syst 226:107134. https://doi.org/10.1016/j.knosys.2021.107134 Birjali M, Kasri M, Beni-Hssane A (2021) A comprehensive survey on sentiment analysis: Approaches, challenges and trends. Knowl Based Syst 226. https://doi.org/10.1016/j.knosys.2021.107134 Chandrasekaran R, Mehta V, Valkunde T, Moustakas E (2020) Topics, Trends, and Sentiments of Tweets about the COVID-19 Pandemic: Temporal Infoveillance Study. J Med Internet Res 22. https://doi.org/10.2196/22624 Xue J, Chen J, Chen C, Zheng C, Li S, Zhu T (2020) Public discourse and sentiment during the COVID 19 pandemic: Using latent dirichlet allocation for topic modeling on twitter. PLoS ONE 15. https://doi.org/10.1371/journal.pone.0239441 Boon-Itt S, Skunkan Y (2020) Public perception of the COVID-19 pandemic on twitter: Sentiment analysis and topic modeling study. JMIR Public Health Surveill 6. https://doi.org/10.2196/21978 Yin H, Song X, Yang S, Li J (2022) Sentiment analysis and topic modeling for COVID-19 vaccine discussions. World Wide Web 25. https://doi.org/10.1007/s11280-022-01029-y Qorib M, Oladunni T, Denis M, Ososanya E, Cotae P (2023) Covid-19 vaccine hesitancy: Text mining, sentiment analysis and machine learning on COVID-19 vaccination Twitter dataset. Expert Syst Appl 212. https://doi.org/10.1016/j.eswa.2022.118715 Abdul-Mageed M, Kübler S, Diab M (2012) SAMAR: a system for subjectivity and sentiment analysis of Arabic social media, in: 12 Proceedings of the 3rd Workshop in Computational Approaches to Subjectivity and Sentiment Analysis Zarra T, Chiheb R, Moumen R, Faizi R, Afia AE (2017) Topic and sentiment model applied to the colloquial Arabic: A case study of Maghrebi Arabic, in: ACM International Conference Proceeding Series, https://doi.org/10.1145/3128128.3128155 Shelke N, Deshpande S, Thakare V (2017) Domain independent approach for aspect oriented sentiment analysis for product reviews. Adv Intell Syst Comput. https://doi.org/10.1007/978-981-10-3156-4_69 Madani Y, Erritali M, Bouikhalene B (2023) A new sentiment analysis method to detect and Analyse sentiments of Covid-19 moroccan tweets using a recommender approach. Multimed Tools Appl. https://doi.org/10.1007/s11042-023-14514-x Hankar M, Birjali M, El-Ansari A, Beni-Hssane A (2022) COVID-19 Impact Sentiment Analysis on a Topic-based Level. J ICT Stand 10. https://doi.org/10.13052/jicts2245-800X.1027 Hankar M, Birjali M, El-Ansari A, Beni-Hssane A (2022) Arabic Topic Modeling-Based Sentiment Analysis on COVID-19 Feedback Comments, in: Lecture Notes in Networks and Systems, https://doi.org/10.1007/978-3-030-91738-8_9 Hegazi MO, Al-Dossari Y, Al-Yahy A, Al-Sumari A, Hilal A (2021) Preprocessing Arabic text on social media, Heliyon. 7 https://doi.org/10.1016/j.heliyon.2021.e06191 Abu Farha I, Magdy W (2021) A comparative study of effective approaches for Arabic sentiment analysis. Inf Process Manag 58. https://doi.org/10.1016/j.ipm.2020.102438 Imran M, Qazi U, Ofli F (2022) TBCOV: Two Billion Multilingual COVID-19 Tweets with Sentiment, Entity, Geo, and Gender Labels, Data (Basel). 7 https://doi.org/10.3390/data7010008 Blei DM, Ng AY, Jordan MI (2003) Latent Dirichlet allocation. J Mach Learn Res 3. https://doi.org/10.1016/b978-0-12-411519-4.00006-9 Hoyer PO (2004) Non-negative matrix factorization with sparseness constraints. J Mach Learn Res. 5 Rudkowsky E, Haselmayer M, Wastian M, Jenny M, Emrich Š, Sedlmair M (2018) More than Bags of Words: Sentiment Analysis with Word Embeddings. Commun Methods Meas 12. https://doi.org/10.1080/19312458.2018.1455817 Kasri M, Birjali M, Beni-Hssane A (2019) A comparison of features extraction methods for Arabic sentiment analysis, in: Proceedings of the 4th International Conference on Big Data and Internet of Things, ACM, New York, NY, USA, : pp. 1–6. https://doi.org/10.1145/3372938.3372998 Kasri M, Birjali M, El Ansari A, Beni-Hssane A (2022) Enhanced Word Embeddings with Sentiment Contextualized Vectors for Sentiment Analysis, in: : pp. 77–86. https://doi.org/10.1007/978-3-030-91738-8_8 Kasri M, Birjali M, Nabil M, Beni-Hssane A, El-Ansari A, Fissaoui ME (2022) Refining Word Embeddings with Sentiment Information for Sentiment Analysis. J ICT Stand. https://doi.org/10.13052/jicts2245-800X.1031 Devlin J, Chang MW, Lee K, Toutanova K (2019) BERT: Pre-training of deep bidirectional transformers for language understanding, in: NAACL HLT 2019–2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies - Proceedings of the Conference Le Q, Mikolov T (2014) Distributed representations of sentences and documents, in: 31st International Conference on Machine Learning, ICML 2014 Egger R, Yu J, Topic Modeling Comparison Between A (2022) Top2Vec, and BERTopic to Demystify Twitter Posts. Front Sociol 7. https://doi.org/10.3389/fsoc.2022.886498 Abuzayed A, Al-Khalifa H (2021) BERT for Arabic Topic Modeling: An Experimental Study on BERTopic Technique. Procedia Comput Sci 189:191–194. https://doi.org/10.1016/J.PROCS.2021.05.096 Antoun W, Baly F, Hajj H (2021) AraBERT: Transformer-based Model for Arabic Language Understanding Obeid O, Zalmout N, Khalifa S, Taji D, Oudah M, Alhafni B, Inoue G, Eryani F, Erdmann A, Habash N (2020) CAMeL tools: An open-source python toolkit for arabic natural language processing, in: LREC 2020–12th International Conference on Language Resources and Evaluation, Conference Proceedings Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5435843","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":376930405,"identity":"369d0cdb-706e-499c-8d57-9d6912ba1ad8","order_by":0,"name":"Mustapha Hankar","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABB0lEQVRIiWNgGAWjYBACAwYGZgSPsYFBDkQfeECcFmawFmOwlgRStCQ2gNj4tJiznz1s8HMPQ7TB+fMHPxfusEufH3b4IdAWOzndBuxaLHvykhN7njHkbriRzCw980xy7sbbaQZALcnGZgdwOOxAjvEBngMgLcwM0rxtzLkbZyeAtBxI3IZLy/k3xgf/gLScP8z8m7etPt1wdvoH/Fpu5Bgng205kMwGtOVwgrx0DgFbbrwxNpY5IJE780aymfXMtuOGG6RzCg4kGODxy/kcY8k3B2xy+84ffHy7sK1aXn52+uYPHyrs5HBpgQIJMMkMCRAwiVc5AoC1yDcQqXoUjIJRMApGDAAAmVBlvjNpGYYAAAAASUVORK5CYII=","orcid":"","institution":"Chouaib Doukkali University","correspondingAuthor":true,"prefix":"","firstName":"Mustapha","middleName":"","lastName":"Hankar","suffix":""},{"id":376930406,"identity":"a4a50781-b07e-4992-b70b-dc7f1d2b9a72","order_by":1,"name":"Mohammed Kasri","email":"","orcid":"","institution":"Ibn Zohr University","correspondingAuthor":false,"prefix":"","firstName":"Mohammed","middleName":"","lastName":"Kasri","suffix":""},{"id":376930407,"identity":"2971f131-3d1b-449b-9410-d2223ca4c6d8","order_by":2,"name":"Toufik Mzili","email":"","orcid":"","institution":"Chouaib Doukkali University","correspondingAuthor":false,"prefix":"","firstName":"Toufik","middleName":"","lastName":"Mzili","suffix":""},{"id":376930408,"identity":"31f74e8a-fde9-4f3e-b5e1-b7c8eeea5e7a","order_by":3,"name":"Abderrahim Beni-Hssane","email":"","orcid":"","institution":"Chouaib Doukkali University","correspondingAuthor":false,"prefix":"","firstName":"Abderrahim","middleName":"","lastName":"Beni-Hssane","suffix":""}],"badges":[],"createdAt":"2024-11-12 03:42:12","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-5435843/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5435843/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":68887110,"identity":"bc98541c-3b82-4e74-8f37-dfb0fd90d60f","added_by":"auto","created_at":"2024-11-13 06:50:26","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":33612,"visible":true,"origin":"","legend":"\u003cp\u003eArabic text preprocessing pipeline\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/94c7f6fe12ed42305667d93b.jpg"},{"id":68888134,"identity":"84860809-3453-4f29-b384-df21db39586e","added_by":"auto","created_at":"2024-11-13 06:58:26","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":61843,"visible":true,"origin":"","legend":"\u003cp\u003eTopics distribution of document clusters for each year\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/6e00b612f7c9d74833a12021.jpg"},{"id":68888366,"identity":"b0da1119-1ea7-458c-8705-498d0379a367","added_by":"auto","created_at":"2024-11-13 07:06:27","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":106191,"visible":true,"origin":"","legend":"\u003cp\u003eThe proposed approach workflow\u003c/p\u003e","description":"","filename":"figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/2429bfa67755a53115af61ef.jpg"},{"id":68887107,"identity":"19a13b5c-f030-4e25-b218-15b6dd802b73","added_by":"auto","created_at":"2024-11-13 06:50:26","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":13590,"visible":true,"origin":"","legend":"\u003cp\u003eTopic-based sentiments in 2020\u003c/p\u003e","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/cfbfbd687ebfc60209b0a9bc.png"},{"id":68889214,"identity":"e5e29356-a8e7-46f5-8166-086b60dafe11","added_by":"auto","created_at":"2024-11-13 07:14:26","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":11622,"visible":true,"origin":"","legend":"\u003cp\u003eTopic-based sentiments in 2021\u003c/p\u003e","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/3f4a2082cfcf793827941715.png"},{"id":68888365,"identity":"59a16cd9-90db-42d8-b783-f5c94f5d1222","added_by":"auto","created_at":"2024-11-13 07:06:26","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":9589,"visible":true,"origin":"","legend":"\u003cp\u003eTopic-based sentiments in 2022\u003c/p\u003e","description":"","filename":"Figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/5b6ed8391a2b181937478613.png"},{"id":68889336,"identity":"92e1c14b-df88-4018-a92c-c909d6776717","added_by":"auto","created_at":"2024-11-13 07:22:26","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":42474,"visible":true,"origin":"","legend":"\u003cp\u003eSentiment variation in 2020\u003c/p\u003e","description":"","filename":"Figure7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/d829f7ca0054300649db90b4.jpg"},{"id":68888139,"identity":"ddebe478-a1a2-4d76-b9c5-bc0032246a6e","added_by":"auto","created_at":"2024-11-13 06:58:27","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":80632,"visible":true,"origin":"","legend":"\u003cp\u003eSentiment variation in 2021\u003c/p\u003e","description":"","filename":"Figure8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/382e4cfa016d3f4610cf5874.jpg"},{"id":68888136,"identity":"157af033-6bf6-40d0-a4ab-8d6825cbf94f","added_by":"auto","created_at":"2024-11-13 06:58:26","extension":"jpg","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":87185,"visible":true,"origin":"","legend":"\u003cp\u003eSentiment variation in 2022\u003c/p\u003e","description":"","filename":"Figure9.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/24baeedef9b76c5ce9c37cf4.jpg"},{"id":68887118,"identity":"be9c190e-63ae-45d9-aa34-8e695a195715","added_by":"auto","created_at":"2024-11-13 06:50:27","extension":"jpg","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":74045,"visible":true,"origin":"","legend":"\u003cp\u003eMonthly COVID-19 cases and fatalities in Morocco (2020)\u003c/p\u003e","description":"","filename":"Figure10.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/43b53ed94063c7ed03ebddfc.jpg"},{"id":68887117,"identity":"71b5bd3c-3906-4b45-a5fd-4002f6badba5","added_by":"auto","created_at":"2024-11-13 06:50:27","extension":"jpg","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":76401,"visible":true,"origin":"","legend":"\u003cp\u003eMonthly COVID-19 cases and fatalities in Morocco (2021)\u003c/p\u003e","description":"","filename":"Figure11.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/fc2b26c45e77906c744ca837.jpg"},{"id":68887114,"identity":"5235faa6-702a-44fc-b8a4-c6cd215f4f6a","added_by":"auto","created_at":"2024-11-13 06:50:27","extension":"jpg","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":67442,"visible":true,"origin":"","legend":"\u003cp\u003eTopics distribution of document clusters for each year\u003c/p\u003e","description":"","filename":"Figure12.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/67f700a12ccd8dca2058b21f.jpg"},{"id":68889338,"identity":"61d13345-7264-4a6e-8295-4fcdafa532a0","added_by":"auto","created_at":"2024-11-13 07:22:32","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1235724,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5435843/v1/1bee936d-bdd8-4d3f-a9dc-9b26f302a04a.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eInvestigating the various impacts of COVID-19 using Sentiment Analysis and Topic Modeling over three years\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003e \u003cspan fontcategory=\"NonProportional\" class=\"\" name=\"Emphasis\"\u003ePlease have a look at courier new font provided for text in article.\u003c/span\u003e\u003c/p\u003e \u003cp\u003eThe new coronavirus, also called SARS-CoV-2, has created a calamitous situation around the globe by causing millions of deaths, and infections and putting other lives at stake. The pandemic has severely affected and challenged public healthcare systems, governments, and societies in an unimaginable way. Moreover, its impacts still go on and will remain for many years to come[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. COVID-19 outbreak had a negative impact on every aspect of life including public health, economy, education, travel, social life, labor market, etc.[\u003cspan additionalcitationids=\"CR4 CR5\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. As the pandemic continues to spread exponentially and new variants emerge, it is clear that the world will be facing a long-term crisis and disruption mostly on the economic level.\u003c/p\u003e \u003cp\u003eIn the periods of quarantine and stay-in-home measures, people have suffered from mental health disorders such as stress, anxiety, depression, feelings of isolation, worry, etc. Many conducted studies[\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] showed that protective measures which have been imposed by governments such as travel restrictions, lockdowns, and economic shutdowns have exposed people to psychological disorders while being isolated in their places, socially distanced from one another, and living at a constant risk of losing their work. These effects include symptoms of post-traumatic stress disorder, long quarantine suppress symptoms, and many others. Social media outlets like Facebook and Twitter, news platforms, blogs, and forums, during the periods of the pandemic have been experiencing reactions, feelings, emotions, and thoughts of people towards the COVID-19 pandemic and its impacts [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eNatural language processing (NLP) techniques are broadly applied to retrieve and analyze relevant information from these huge amounts of generated content. For instance, topic models are used in this context to extract the main topics related to the outbreak. Sentiment analysis is another NLP technique that is often applied to identify sentiments and emotions in a given text as a way to investigate the effects of COVID-19 on people\u0026rsquo;s normal lives. Sentiment analysis is widely applied across various domains like business intelligence to analyze customers feedback regarding a service or a product [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], and social media monitoring to analyze people\u0026rsquo;s opinions about public policies or certain figures[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Sentiment analysis approaches are mostly categorized into three main levels: document level, sentence level, and aspect level [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. In document level, the whole document is classified as whether negative, neutral, or positive. In sentence level, subjective sentences are filtered out before classifying their polarity. The aspect-level approach focuses on performing sentiment analysis regarding a specific aspect [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn this paper, we performed a topic-based sentiment analysis on a corpus of collected comments related to COVID-19 outbreak. The dataset that is made available for this study consists of comments scraped from the online newspaper Hespress\u003ca class=\"FNLink\" href=\"#Fn1\" id=\"#FNLinkFn1\"\u003e†\u003c/a\u003e written in both modern standard Arabic (MSA) and Moroccan dialect (MD). Our approach follows a process of two steps: topic extraction and sentiment analysis. For topic modeling, we applied an embedding-based topic model, namely BERTopic, to retrieve topics from the corpus and generated document clusters based on the extracted themes. Second, a topic-based sentiment analysis is carried out to identify polarities within the texts and classify them into positive, neutral, or negative. Finally, a time series analysis is performed on a monthly basis to track changes in sentiments in the period of three years of the pandemic. The visualization of these changes shows how people's reactions and sentiments evolve over time regarding the COVID-19 situation. Time series sentiment analysis can relate events like quarantine, COVID-19 spikes of infections and deaths, travel restrictions, etc.; to the changes in sentiments and emotions of people. For example, our study showed that negative sentiments of fear, anxiety, and depression are increased during quarantine periods and when a new wave of COVID-19 variant emerges and spreads. Negative sentiments also increased regarding the vaccine topic indicating that people hold suspicious sentiments towards vaccination campaigns.\u003c/p\u003e \u003cp\u003eThe key contributions to this paper are: (i) collecting a real dataset containing thousands of comments written in Arabic, (ii) designing a semi-supervised model that integrates both topic modelling and sentiments analysis techniques to examine the effects of COVID-19 in Morocco, and (iii) tracking the changes of sentiment analysis results from March 2020 until the end of December 2022 in correlation to the epidemiological situation. The rest of the article is organized as follows: The second section is dedicated to review and discuss related works to the study. In section 3, we introduce the methods and the materials used to perform this research. Results are presented and discussed in section 4. Finally, we summarize our paper and discuss some feature works in the section 5.\u003c/p\u003e\u003cp\u003e[\u003cstrong style='text-align: left;color: rgb(0, 29, 53);background-color: rgb(255, 255, 255);font-size: 16px;font-family: \"'\u003e\u0026dagger;\u003c/strong\u003e] \u003cstrong\u003eHesspress\u003c/strong\u003e news website (accessed 26/5/2024): https://www.hespress.com/\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e"},{"header":"2. Related Works","content":"\u003cp\u003eIn recent years, academic researchers have shown growing interest in the field of NLP in order to design advanced techniques for processing human languages. However, Arabic NLP is still facing many challenges as a result of some specific and intrinsic features of the language. This section will present and examine some relevant research that relates to our study.\u003c/p\u003e \u003cp\u003eChandrasekaran et al.[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] analyzed trends, topics, and sentiments in a dataset that contains tweets about COVID-19 outbreak. They first applied Latent Dirichlet Allocation (LDA) to extract topics within the dataset and then implemented a dictionary-based method, namely valence aware dictionary and sentiment reasoner (VADER), to compute sentiment scores. In another work, Xue et al. [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] analyzed COVID-19 tweets using LDA and machine learning-based sentiment analysis to examine public discourse and reactions of tweeters to the pandemic. Their results showed that people expressed feelings of fear and threat regarding the unknown nature of the coronavirus virus. Boon-Itt and Skunkan [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] conducted a study by utilizing LDA and sentiment analysis techniques to explore and identify topic discussions in a dataset of collected Tweets over time. In their experiment results, they claimed that people expressed negative feelings towards the COVID-19 outbreak.\u003c/p\u003e \u003cp\u003eYin et al. performed [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] an in-depth analysis of social media posts and discussions related to COVID-19 vaccines on Twitter. In the first step, the authors used LDA to extract the main topics about various vaccines and then applied VADER model to compute sentiment polarities. Reported results showed that people feel positively confident to take vaccines. Qorib et al [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] conducted a study to examine COVID-19 vaccine hesitancy among people through their social media posts. They applied several machine learning algorithms combined with different vectorization methods (TF-IDF, Doc2vec, and BoW) to predict hesitancy regarding vaccines. In their experiment results, they suggest that the combination of TextBlob, TF-IDF, and LinearSVC outperformed other models in sentiment classification. They also concluded that people are feeling optimistic about taking vaccines and hesitancy decreases over time.\u003c/p\u003e \u003cp\u003eAbdul-Mageed et al. [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], developed a hybrid model for subjectivity detection and sentiment analysis. The model first detects subjectivity within a text and then classifies its polarity into neutral, positive, or negative. They tested the model on a dataset collected from social media outlets like Twitter and other web platforms such as Wikipedia Talk Pages, mini blogs, chat apps, and web forums. The model utilizes a machine learning algorithm (SVMlight) to classify sentiments. However, they faced challenging and complex issues in their experiments due to the specific characteristics of the Arabic language. In [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], Zarra et al. introduced a semi-supervised model that integrates topic modeling with sentiment analysis to conduct aspect-oriented opinion mining. They used an unsupervised method to retrieve the main themes within Facebook comments written in colloquial Maghreb and then applied a supervised technique to compute sentiment polarity. Shelke et al.[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] designed an aspect-oriented model to conduct sentiment analysis on a dataset of product reviews. They used SentiWordNet lexicon and some specific features of reviews to identify their polarity. Madani et al.[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] designed a sentiment analysis recommender to classify tweet as positive, negative, or neutral. The dataset they used for the study contains COVID-19 related tweets from Morocco, collected between March 2020 and October 2020. Their experimental results showed that their proposed model reached 86% accuracy outperforming baseline machine algorithms. They found out that changes in sentiments over time are affected by the COVID-19 epidemic situation. In our previous works [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], we used LDA to extract topics from a collected dataset that contains more than 20,000 comments from Hespress, and then we applied a pretrained transformer model to classify sentiments into negative, positive, and neutral. The findings showed that negative sentiments were high regarding all extracted topics in the first year of the pandemic.\u003c/p\u003e"},{"header":"3. Proposed Approach","content":"\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\n \u003ch2\u003e3.1 Datasets\u003c/h2\u003e\n \u003cdiv id=\"Sec5\" class=\"Section3\"\u003e\n \u003ch2\u003e3.1.1 COVID-19 Time Series\u003c/h2\u003e\n \u003cp\u003eMany sources such as the Moroccan ministry of health, WHO, and the university of Johns Hopkins contributed to the collection of this dataset. Since January 22, 2020, the team at the at Johns Hopkins Center for Systems Science and Engineering (CSSE) has been responsible for recording and updating COVID-19 data from across the globe. They have been cleansing, transforming, and normalizing data to make it easier for analysis and further processing. They arranged dates and consolidated several files to transform data into normalized time series. The dataset is made available in a GitHub data repository as a CSV file and can be accessed via this link\u003ca id=\"#FNLinkFn2\" class=\"FNLink\" href=\"#Fn2\"\u003e\u0026Dagger;\u003c/a\u003e. The dataset contains six columns: the dates that confirmed cases or fatalities were recorded, cumulative confirmed cases, cumulative deaths, recovered cases, Region/Country, and finally Province/state. We filtered out the time series based on the country column to retrieve cases and fatalities based in Morocco from March 2, 2020, the date in which the first case appeared, to December 31, 2022. We transformed the dataset into a time series of daily confirmed cases and daily fatalities indexed by DateTime.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e\n \u003ch2\u003e3.1.2 Hespress Comments\u003c/h2\u003e\n \u003cp\u003eThe dataset contains more than 20,000 comments sourced from the prominent news outlet, Hespress. These comments are written in both Modern Standard Arabic (MSA) and Moroccan Dialect (MD). For data collection, we developed a Python crawler using the Selenium library to extract comments from news articles related to the COVID-19 pandemic across various domains, including health, economy, politics, society, and vaccines. The scraping process was conducted during different time frames between March 2020 and December 2022 to ensure comprehensive coverage of the COVID-19 timeline. Comments in languages other than Arabic, such as English, French, or Spanish, were filtered out to retain only Arabic comments. The resulting dataset is stored as a CSV file, containing columns for the publication date, the user\u0026apos;s name, and the comment text. The dataset can be accessed on GitHub via this link\u003ca id=\"#FNLinkFn3\" class=\"FNLink\" href=\"#Fn3\"\u003e\u0026sect;\u003c/a\u003e.\u003c/p\u003e\n \u003cp\u003eArabic language processing is still facing numerous challenges due to some inherit characteristics of the language itself like metaphor, diglossia, ambiguity, and various other difficulties related to Arabic morphology [\u003cspan class=\"CitationRef\"\u003e27\u003c/span\u003e]. Moreover, researchers are frequently encountered with a mixture of three versions of the Arabic script on the web: Modern Standard Arabic, Classical Arabic, and Dialectical Arabic. The differentiations in these scripts make processing and understanding tasks more difficult. Arabic language has a different structure and morphology compared to Latin-based languages. For instance, Arabic script is written from right to left and has 28 different letters. The Arabic letters can have diacritics such as hamza(همزة:ء), sukun(سكون: ْ), fatha(فتحة: َ), kasra(كسرة: ِ), tanwin(تنوين: ً), etc. These diacritics can directly affect the semantics of words in the text and increases ambiguity to capture its meaning.\u003c/p\u003e\n \u003cp\u003eThe morphology of Arabic letters can change according to their location (initial, medial, final, or isolated) in the text which makes the processing task even more challenging. Cleaning texts in Arabic includes text normalization and removing punctuations, diacritics, numbers, links, and elongations [\u003cspan class=\"CitationRef\"\u003e28\u003c/span\u003e]. The preprocessing task starts with tokenization to split the text into tokens, stop words removal to eliminate non-informative words such as fi (في), min (من), ala (على), etc., and stemming to transform words into their base form by removing suffixes and prefixes [\u003cspan class=\"CitationRef\"\u003e27\u003c/span\u003e]. Table 1 below shows a sample of texts in the dataset before and after preprocessing and Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates the main steps to preprocess Arabic text.\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;Table 1. Examples of comments before and after preprocessing.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003ctable id=\"Taba\" border=\"1\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRaw text\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePreprocessed text\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eكلنا متضررين و لكن أحسن حاجة هي تمديد الحجر الصحي مع مراعاة الظروف الإجتماعية لينا كاملين. و الله يحد الباس.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eمتضررين احسن حاجة تمديد حجر صحي مراعاة ظروف اجتماعية كاملين الله يحد باس\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eالدار البيضاء هي اكثر المدن كثافة، هي بمثابة مغرب مصغر وهي قلب الاقتصاد الوطني .اظن انه يجب تخفيف الحجر على باقي الجهات التي لم تعد تسجل اصابات عديدة والتركيز على البيضاء و خاصة الشركات اما انتظار تسجيل 0 اصابة للرفع فهو ضرب من. ضروب المستحيل\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eالدار البيضاء مدن كثافة بمثابة مغرب مصغر قلب اقتصاد وطني اظن يجب تخفيف حجر باقي جهات لم تعد تسجل اصابات عديدة تركيز شركات انتظار تسجيل اصابة رفع ضرب ضروب مستحيل\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eبالنسبة للناس اللي ما خداوش جرعة اللقاح راه ما عندهمش اي نية للسفر سواء داخل المغرب او خارجه فاغلب الناس المهتمين بالسفر تلقحو هادشي راه ماشي معقول لا حول ولا قوة إلا بالله\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eناس ماخداوش جرعة لقاح ما عندهمش نية سفر سواء داخل المغرب خارج اغلب ناس مهتمين سفر تلقحو ماشي معقول\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003eArabic text may include Arabic numerals, Arabic-specific characters, or mixed Arabic words with words in languages. Handling these challenges involves making decisions based on the specific requirements of the sentiment analysis task. For example, we might choose to replace Arabic numerals with their written forms or decide how to handle mixed language text and code-switching. Arabic text may contain spelling errors or typos. Applying spell checking and correction techniques can improve the accuracy of subsequent analysis steps. Arabic-specific spell checkers or general-purpose spell checkers with Arabic language support can be used. Additional preprocessing steps may be required for the Arabic language. For example, for aspect-based sentiment analysis, we may try identifying and extracting aspect terms or entities from the text using named entity recognition or part-of-speech tagging.\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e\n \u003ch2\u003e3.1.3 TBCOV Tweets\u003c/h2\u003e\n \u003cp\u003eTBCOV, also known as two billion COVID-19, is a large-size Twitter collection [\u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e] that contains more than 2B multilingual COVID-19-related tweets. Specifically, TBCOV contains over two billion tweets using more than 800 keywords from different languages. The collection was gathered within a period of 14 months beginning from February 1, 2020, until March 31, 2021. The tweets are written in 67 different languages (including Arabic) and posted by more than 87\u0026nbsp;million unique users on Twitter across 218 countries all over the world. Due to the largeness of the original multilingual dataset, the collectors have offered different filters based on location, language, and time window on their website\u003ca id=\"#FNLinkFn4\" class=\"FNLink\" href=\"#Fn4\"\u003e**\u003c/a\u003e. We used language and location filters to retrieve more than 40,000 Arabic-written tweets that were published in Morocco during the same period of collection. The Arabic tweets are cleaned, preprocessed, and prepared for topic modeling and sentiment analysis using NLP tools.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\n \u003ch2\u003e3.2 Topic Extraction using BERTopic\u003c/h2\u003e\n \u003cp\u003eTopic modeling is a text-mining method that is commonly used to discover the hidden themes within a corpus of textual documents. Topic models represent documents in the collection as a mixture of different topic, and in turn, each topic is considered as a distribution of various words weighted by their respective scores. Topic models are widely applied in several domains like information retrieval, text classification, sentiment analysis, and recommendation systems, etc. Classic topic models like LDA[\u003cspan class=\"CitationRef\"\u003e30\u003c/span\u003e] and non-negative matrix factorization (NMF)[\u003cspan class=\"CitationRef\"\u003e31\u003c/span\u003e] are limited to representing documents as a bag of words (BoW) which leads to ignoring the order of words, semantic relationships, and contextuality between words [\u003cspan class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e33\u003c/span\u003e]. In response to this issue, word embedding models have emerged in NLP field to address the problems of BoW [\u003cspan class=\"CitationRef\"\u003e34\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e35\u003c/span\u003e]. For instance, Bidirectional Encoder Representations from Transformers (BERT)[\u003cspan class=\"CitationRef\"\u003e36\u003c/span\u003e] have shown promising results in generating contextual text representations that capture the semantic structures within sentences and word vectors. Modern topic models take great advantage of the of word embeddings to build coherent models powered with centroid-based techniques. For example, BERTopic generates representations of topics in documents in a way that each topic is assigned to a cluster of documents. The term frequency-inverted document frequency (TF-IDF) is computed by multiplying term frequency \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:t{f}_{t,d}\\)\u003c/span\u003e\u003c/span\u003e of a word \u003cem\u003et\u003c/em\u003e in document \u003cem\u003ed\u003c/em\u003e by the inverse document frequency (IDF). The value of IDF is derived by calculating the logarithm of the corpus size N divided by the total number of documents containing the term t. The formula is shown in Eq. (\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e) below:\u003c/p\u003e\n \u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\n \u003cdiv id=\"FileID_Equ1\" class=\"mathdisplay\"\u003e$$\\:{\\text{t}\\text{f}\\text{i}\\text{d}\\text{f}}_{t,d}={\\text{t}\\text{f}}_{t,d}\\times\\:\\text{l}\\text{o}\\text{g}\\left(\\frac{\\text{N}}{{\\text{d}\\text{f}}_{t}}\\right)$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003eThis TF-IDF formula is adjusted to measure the importance of a term to a topic instead of a document. The Eq.\u0026nbsp;(\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e) measures the class-based TF-IDF of a term in a given class:\u003c/p\u003e\n \u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\n \u003cdiv id=\"FileID_Equ2\" class=\"mathdisplay\"\u003e$$\\:{W}_{t,c}=t{f}_{t,c}\\cdot\\:\\text{l}\\text{o}\\text{g}\\left(1+\\frac{A}{t{f}_{t}}\\right)$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:t{f}_{t,c}\\)\u003c/span\u003e\u003c/span\u003e measures the frequency of a term \u003cem\u003et\u003c/em\u003e in class \u003cem\u003ec\u003c/em\u003e representing a set of documents grouped into one document known as a cluster. This term frequency is then multiplied by the inverse class frequency (ICF) to assess the importance of a term within a class. The value of ICF is obtained by computing the logarithm of the average number of terms in each class A divided by the frequency of the term \u003cem\u003et\u003c/em\u003e across classes. The one inside the logarithm is added to the division to ensure the output score values remain positive. The class-based TF-IDF assesses the relevance of terms in document clusters allowing to extract topic-term distributions for each cluster.\u003c/p\u003e\n \u003cp\u003eIn this paper, we applied a word embedding-based technique, namely BERTopic, to extract topics from Hespress comments and tweets to generate document clusters. Afterward, each document cluster is assigned to each extracted topic based on the aspect representations. As for that, we used BERTopic due to its good results in generating coherent and diverse topics when compared to the performance of conventional topic models such as LDA[\u003cspan class=\"CitationRef\"\u003e30\u003c/span\u003e], NMF[\u003cspan class=\"CitationRef\"\u003e31\u003c/span\u003e], Doc2vec[\u003cspan class=\"CitationRef\"\u003e37\u003c/span\u003e], and Top2vec[\u003cspan class=\"CitationRef\"\u003e38\u003c/span\u003e] in various Arabic datasets[\u003cspan class=\"CitationRef\"\u003e39\u003c/span\u003e] including Hespress comments dataset. Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e shows experimental results of four topic models in terms of topic coherence (TC) and topic diversity (TD). TC measures the rate of semantic similarity between the most important words in a topic while TD measures the degree of diversity among topics.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"char\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"char\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003ePerformance of topic models leveraged with topic coherence and topic diversity across different topic numbers (K).\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth rowspan=\"2\" align=\"left\"\u003e\n \u003cp\u003eTopic Model\u003c/p\u003e\n \u003c/th\u003e\n \u003cth colspan=\"2\" align=\"left\"\u003e\n \u003cp\u003eK\u0026thinsp;=\u0026thinsp;6\u003c/p\u003e\n \u003c/th\u003e\n \u003cth colspan=\"2\" align=\"left\"\u003e\n \u003cp\u003eK\u0026thinsp;=\u0026thinsp;10\u003c/p\u003e\n \u003c/th\u003e\n \u003cth colspan=\"2\" align=\"left\"\u003e\n \u003cp\u003eK\u0026thinsp;=\u0026thinsp;15\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTD\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTD\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTD\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLDA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.021\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.029\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.33\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNMF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.158\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.196\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTopic2Vec\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.093\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.099\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.154\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.672\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBERTopic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.211\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.992\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003eFrom Table 2, we can see that BERTopic outperforms other topic models with high scores of topic coherence and topic diversity. \u0026nbsp;The results shown in the table were recorded after five iterations of training across three different topic numbers (K=6, K=10, K=15) to avoid overfitting. \u0026nbsp;The word embeddings of texts were generated using an Arabic pre-trained word embedding model, namely AraBERT [40]. Afterward, we trained the BERTtopic model on top of the constructed word embeddings to extract the themes that are related to the COVID-19 pandemic. \u0026nbsp;The extraction of topics is performed in the Algorithm 1.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\n \u003ch2\u003e3.3 Sentiment Classification using CAMeL\u003c/h2\u003e\n \u003cp\u003eSentiment analysis (SA) is an NLP technique which is often used to identify sentiments within subjective text and classify its polarity into negative, neutral, or positive. Sentiment analysis techniques are typically categorized into three primary types: lexicon-based techniques, machine learning techniques, and hybrid approaches[\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e]. Machine learning techniques combine statistical and probabilistic algorithms with specific linguistic characteristics to identify sentiments in a given text. However, Arabic sentiment analysis (ASA) is particularly facing numerous challenges because of various language-specific characteristics. To address these challenges, researchers have contributed to developing various tools and resources, such as CAMeL [\u003cspan class=\"CitationRef\"\u003e41\u003c/span\u003e]. This open-source toolkit is used to perform various Arabic NLP tasks including sentiment analysis, part-of-speech tagging, and name entity recognition. CAMeL sentiment analyzer is a pre-trained model that is fine-tuned on AraBERT and multilingual BERT word embeddings to detect sentiments in Arabic texts. The classifier was trained and evaluated on different datasets including the Arabic Speech-Act dialectical dataset and the ArSAS tweet sentiment corpus. Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e shows the evaluation results of the CAMeL sentiment analyzer over three datasets.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"char\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"char\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"char\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eCAMeL sentiment analyzer accuracy using AraBERT and mBERT compared to Mazajak over three benchmark datasets [\u003cspan class=\"CitationRef\"\u003e41\u003c/span\u003e].\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eCAMeL(AraBERT)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eCAMeL(mBERT)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMazajak\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eArSAS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eASTD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSemEval\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.63\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\n \u003ch2\u003e3.4 Topic-based Sentiment Analysis\u003c/h2\u003e\n \u003cp\u003eThe conventional models employed for Arabic sentiment analysis often focus on computing text polarity on the document level while ignoring the aspect/topic level that holds an opinion towards a specific entity within the text. In our approach, we consider that subjective texts express multiple topics and each one of them has its own sentiment. Given this, we chose to perform sentiment analysis on a topic-based level to capture people\u0026apos;s opinions and examine their reactions regarding many different COVID-19 related aspects or topics. Our approach follows a process of two steps: we first extract topics from the corpus of documents using BERTopic and then generate word distributions of extracted topics based on word scores representing each specific topic. Afterward, document clusters are constructed and assigned to map each extracted topic with a corpus of documents using the c-TF-IDF formula. Table \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e presents the final topics and their word distributions while Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e shows the distribution of document clusters for each topic.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eHigh-scored words for each topic.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTopic\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eHigh-scored topic words\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEconomy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eمحلات | متاجر | شغل | صناعة | معامل | اقتصاد | عمل | اغلاق | شركات | أرباب\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEducation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eدراسة | تعليم | عمومي | عن بعد |حجر صحي | استاذ | امتحان | مدرسة | منصة | تربية | وزارة | تلميذ\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFear\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eجوع | ملل | خطر | موت | بلاء | قلق | خوف | عدوى | شفاء | الله | سقم | انتحار| مصيبة | تشاؤم | صحة\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSupport\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eتمديد | تعويضات | راميد | صندوق | دعم | ازمة | بطالة | ضمان | مساعدات | حجر صحي | داخلية | طوارئ | التزام | دولة | حكومة | مواطن | حظر\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHealth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eتعقيم | اعراض | فقدان | شم | ذوق | تنفس | نقص | اختناق | وقاية | حالات | صحة | تحاليل | اصابات | كمامة\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVaccine\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eجواز | دلتا|لقاح | كوفيد | اوميكرون | سينوفارم | استرازينيكا | فايزر | فيروس | تلقيح | متحور | الغاء | جرعة خطة | مسرحية | كذب | اوهام | مؤامرة | ماسونية | ابادة | تأثير\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eIn the second step of our approach, we transformed constructed document clusters and saved them as data frames and each data frame represents a document cluster of a specific topic. We mention that a text in the corpus can belong to multiple clusters at once and thus it may hold many opinions about different topics. After that, we used CAMeL to classify texts of document clusters into positive, neutral, or negative. The results of the whole process from text representation to topic clustering to sentiment analysis is a refined aspect-based infodemic sentiment analysis. Algorithm 2 represents the performed sentiment classification task.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003cimg src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAA4oAAAHeCAYAAADdDfUwAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAAFiUAABYlAUlSJPAAAGbTSURBVHhe7d3ri13bntf/Vb/nXrr1aRLjPpISFUGPClZCUNgpUREFtT1YORzYidqNCKKeo5V2c9yepN3eHoiXo5UNm53g3qKgCK1W+kGTrkS68UILghXonJhEfHS8/wH1W59R81P5zm/GvKxVVavWqnq/YGSuNS9jjjHmmKnxrXmptYOpCQAAAAAAjf+vmQIAAAAAUBAoAgAAAABaCBQBAAAAAC2jn1H87ne/23wCAAAAACzaImMyrigCAAAAAFoIFAEAAAAALfx5DAAAAABAC1cUAQAAAAAtBIoAAAAAgBYCRQAAAABAC4EiAAAAAKCFQBEAAAAA0LLwQPEnfuInJr/n9/ye5lvds2fPJmtra620LH7wgx9Mvva1r01+9Ed/dPKv/tW/aua25frpe6zLX//rf71ZcrZUDtVFZdL0JMpVO3Zj0qL4WJz0MTir+hxX7Xhp3hha74/9sT92tN1v+22/bfLVV19N/sf/+B/l8yJpnypLdlrHe1HUxscpu46H/q9S+5yVRf7/t+rHGwCApaI/j7EoP/zhD/WnOEp6+fJlM7fup3/6p4/WXXAxe3366adHZdrc3GzmvvP973+/Wt6vf/3rR9spj7P24z/+40fliUnzj2Nvb+/gR37kRw6+/PLLcrwltpmSqQ/82I/9WGveaVLZauU4CbFvn3Tepy2XXe00RMdX66q/+Dhru9jPF0n9KJ+Pp3m8F0Ht+sEHHxzr/wsfDx2vs7SI//9W/XgDALBsFnpF8Z/9s3/WfJpMPvvss+ZT3S//5b+8+bRc/vAf/sOT6eBtMg2GJn/mz/yZZu4h/fb/T/2pP9V8a/tVv+pXNZ/Onq4y/IN/8A+ab22a/5/+039qvs3np37qp8rVnaE6/9pf+2tLWdSei3D9+vXJNJgon6eD1TI9Kct0fGc1a9l1dUp3Bsjf//t//2h7te+/+3f/7qiNF0VXj/7JP/knzbd3TvN4L8If/+N/fPLy5cvm2+x0Hv/7f//vy+e/9bf+VpmelUWcH6t+vAEAWDYLDRT/xt/4G82nw4BkFSm4+aVf+qXJ//yf/3Pye3/v723mTsptqH/gD/yB5tty06Dx+9//vn7lPvnhD384+fEf//FmyaHd3d3m0+w0WPuTf/JPNt/GUXsuyr/5N/+m1Pvb3/52Mwez+i//5b9M/tf/+l/ls27FzhQ8LoqCxO985zvNt/et4vH2bbTHOQ/lH/7Df9h8mpSA8bi/AFoFnN8AAJychQWK+bknDTR1NWkeykvPQMXnXmLScyqRBpNDz+LlPLQPDdj8zIsCQW0X1/F+NCD7fb/v9x0NnsXr5HrL//k//6dckdGzQ1rn3r17zZJDtf2orVwHbec65LqNGQz+lb/yV46COf2mXwP7eFXv8uXLzadDfs5JaRGDTV+xcvvoWOfnQX1cnNSG8Xk5fY7PZek4xPWVMh1Ht6WS8qgFQmOonVxG5RkH7ZHmx7po3dqzr/H4q67uj1nsD7V2k1i22Jfm8WM/9mPv9Qn9MqV2VTEfV7Vv3Fbl0HwnlT/WR2WOx1TbxyBRgZW37Tvetb6j/bhcPga5vHn/pjq472ldbRPXy/vT8phvPP/V37R+vEKqOmo9zR9L+9fV3airD8osZYzG9t8sH2sni/P8f2jfOTB0fmt91UVJ+eh77f9mAAAQHCyIniHSMyR65ka7VZoOJpul7+t63iQ+u6jnAf0cj75r6uelzM/iaf/iZ+Lys3jazvkq6ZmeX/zFXyxl1Hc/V7O9vX20Tiz/0PMxzkdJz/Dp+byYV36GKC7z8lhXJeWp/aqcnqfl84jly8+PxmWxzmOp7by9Uh/V0c8z5eOb2yg+96Q2VVvE7TXVd+t77tX9xPVzHrO0Z8xbx0/7Vpk9z33I3K4qt7hvKqmspuOheS6b1ld9c365r+f8RXlpW8339n6u1imuX+PyxKR9534T1Y6ryqEUj1Hs91qmvh2PWz5vY9/KfbPveOe+o/3E80jz1I6qU2yfvH9to3W1TJ/VdvqcyxL3537Zd/772Cnl4zyGyqzt3CeUcltn85ZxqP9KrT6xbEqxbPqsed7nmHOg63j7HPS6Pp5D/RwAgItuIVcU9Vty3fqk2xKng4lm7uEVgFmv2MTnAnVVTFfEbt26Vb5PBxOt5yD1G2ff4vqn//Sfbk01X79ZtvwMzZs3bya/+Tf/5slP/uRPlu+/8Tf+xjL9Fb/iV5TpcfzFv/gXy1WXmNfP/dzPNZ8OxWXTwVG5YqEy6jfq9lt/628tbapymtpg3qtgMh28lbJF3/rWt8ozmUrTwVYz93T87b/9t4+eq/Lx/RN/4k+U79/4xjdadYvHTO2jttC8O3fulHnKJ/aHrudedWXB/cT968/9uT9XpqrzPO7fv1/KonJNg80y76/9tb92dKVJ+/Sthf/6X//rMv0tv+W3lKl88cUXzafJ5Gd+5meaT4dUz3x7Z+zrf+kv/aUy/d2/+3eX6Z/9s3+2TOVv/s2/eXTlW8/byqy3Cqt/TAOI5tsh7Vv11FWoWv/zcVV7+rj+9t/+20tZtMxiv9dy9e143OI5O6TvOefYd3Q+aj/xPFK5vvnNb5a6/obf8BuauZPJkydPmk+H1Ne0rv5f0/Y6NqJjG6+sxf39kT/yR8r3WNd/8S/+RfPpZOg2fx3f3//7f38z57BOuS9Fs5Rxlv7b5a/+1b/aupMhl83/78mYc6DreOe2Vf/T/3MAAKDfQgJFDdb/wl/4C+XzRx99VKY29FKbSLd4KRDKfs2v+TXNp8PbOm1osNI3OPMASQOSaUDdeh7xNMwT3MVBXPTf//t/bz6No+BFgz4N4jV4yzRY0zOZSnEwfRoc7HTpGujGPhAH9v/8n//z5lO3v/t3/27z6d1gU3XWcc+3783Dwb0G6r/wC79QPv/6X//rj27PrB3H//2//3fz6V2f1jHSbX46XipffA4r9vV8jBSgOUCNgVb+hcAsFAjXfmmg4/f1r3/9vdsPfVwV/GU5+OqjNlyUWuAR///R/0f+pUYMkuz58+fNp2HxeB+XgjgdAx1f/b8Vg7HjvNQmlnGW/ttFgej3vve95ttk8vnnnzefDs/z+EvBMedAF5dFt/D6FloFmQ7qAQBA3UICxX/0j/7R0dULDV48wJDa2wpPytAAtG/fMdg47xys/8t/+S9bVxUWTQPvoUBgTOAX+apHn1kCleP6z//5P5ep2tkv3tAvT/Ssla441vjcEdXn1/26X9d6tk9iHfyMVnx+Ty+gGdO+s9AgXYFTvjqjfeiZXf/yI+5X5Xf5fGwcbK2an//5n28+vXuOUMn+w3/4D82nxfrH//gfl6uh5ivyorY+zh0HNkv/7fPhhx8eXbVXf3C/VtAYfzk35hzoojsi7MGDB+XZV//iBAAAdDv1QFFXFnRraAxA4g9uDTTz1YcuulISfztu8Sri7/ydv7P5tNirD6tKg0YN8L788ssz/w37//t//6/5tFhn2U/U9zXo/af/9J+WWyBr9MsVHR9TeX/X7/pdrZdxxDrs7e2VAXxMOran0b4qm67O6DyOt5WLbnOVvN9cNqVVFP/f0S+/cp0USC2aAiD90qAryJJZ7uIYMqb/9tHPhfiLBr0cR/8n6bb6aMw50EVXHuM+FCz/jt/xOwgWAQAYcOqBom6J021n/m27kp41i8Y8z2J/5+/8nebT4aBCP+x9ZVCDAW4nmo0G9xroazCFyeT//t//23xq63pL4yxvopT4Rlnl6bflDl1Z1/HRQNlXX7RN/IXLIqnc+WqOBvK6rTUO5k/iyhVmo9v8FbTHPpp/uTZ0e/dYs/TfPvFxBJVNv2D4o3/0jzZz3jnOOaD/4+JztWqjv/yX/3LzDQAA1JxqoKiBon57m3/TrhR/aGuQMXZQqd+U/+Iv/mIZ/OiP2//qX/2ry8BBA4j8coN8hSOLt8BeRHrxiF7aEoNEBd6afxa/bVeQ70FgF7+gZawxxziu0/VcmW6zrPXjWa8a+Rk9BVq+NVT7V6A1RMfpZ3/2Z4/aSINd0zNp9m//7b9tPrXpubKT0vWnFlRGl+VX/spfWaZ5v2OuAq0Cv+BKfBvtWdNLbNQvcj/96Z/+6WaNwwBrlpcC1czTf7to23gO6rngrmeh+86BIXqull9kAAAw3qkGigoG/RKbLP/GeOztUApg9KZBPcPiQZAGFjHYsT/4B/9g86nurK7ILAMN9PXbe13djVcfFHjr1rV4q7AGlT/a/A2ysc8Fzat2HKN8G53F2wD/23/7b82nyeQP/aE/1HzqFvuB2iQGyQpq5gma4yD0l37pl8pUV7w9oK4FFrEOka7c+PZsDaAfP35cPseg2m96FT0THPnKu45pHJD7WOb6dV1VjdROXcGGAhHxG4a13/hLGz1DF6l+q0i/tIrHILaH2rQrmD4t2p9fYpPlsh7npTYyS/8dI764JvZlG3MOdNH/KT4f9dm/pPQvMgAAQN2pBIoaJOkHu64Udr2ZUz/s41UQvWQgDqzyYNWDWd0upKuUv+yX/bLyvY8GBR6g+s2WnmrAHIOSPFiOwUYUB0NxG101iYMWDUziADiu6zy68pKuZbV88rZDA30NuHQ1tov/3IgpKNfgXym+IGUMle0//sf/2Hw71DeAjq/Md4DjwEdv2ey6cqHARfXS+n57ovKJgWVXn1I/cAClOqqPaZkCKb36PwbNY/n5PAUPuuqhvhHfKBvPCwWS2tdv+k2/qZnz/jHd2to6CkTevn1bpvG5K9XT55P2p6vCokD30qVLR3WIbyr9qZ/6qTKNf55CdDvhmKt++iWD9hMDTn3X/vNt4PqTHT4/YpCpYxyPUa3fdx03ic8k+5g5777t4ueh86gvn3gXg9tCy/WLr/hnR2r76zrHJV4513qqk9u5RsvVxvplTpf8nF6+c2CWMs7af2t5Rwpkfd53/TJo6BzoOk5666l+DrhP+ypkfOEPAACoODgF00H30R89dop/3Fh/+Dgvd9K2Wre2TGrznaYD5fLHlDPtbzoIKetoqu9Zzkspr1crt8prKrf3o/n+A9K19qgl59W1n1o+tXWVuv6YdPwD4V0pt+GXX35ZtlHS9mN1lc2pq4xqt+kA8KicOq75D3hLbA/ta3t7+2ib6cCw9Qfg+/qU+I+Le/uuftJH26icSiqz8lF+qkv8Y+Ki7y6/pm5Xz1NeandROZSnpspPSWXNxrab5rmfah33CbWZjv3QMVY5VDbtT5+dl+tS26coX+3D6+pz3Jfy8rKhFKnMrrPbpe94x37j1LXvrvnR0PGu7a+WtF6kfDRfebov1NTKqHnW1RZOMmsZZ+m/tbxj+UzHUX2iRuv3nQNDx1v/F7g9Y9kAAEC3Nf0z/eG5MnT1Z+jFCdNBQOtqIc4nvUjGt8BNB5Cj/qYagOWkq4X625Xxja0AAODsnPpbT0/a3/t7f++9t/hlb968aT4BAFbBz/3czxEkAgCwRFYqUNQzJ7qK9CM/8iOTH/7wh7qvqJW+//3vl/W6nosEACwH3fWhZyr17CBvIAUAYPmsVKColyDoJQx6GcEv/MIvNHMP6WUKSl//+tc7X4aA8yW+IOO//tf/2nwCsAr0CIFeHqU/56KX//z5P//nmyUAAGAZrNwzivrts16vrz/h4LfXyebmZnlToP548zxvqcRqic8nRivWnYELS2+81ZtaRW+P5blyAACWy8oFigAAAACA07VyL7MBAAAAAJwuAkUAAAAAQAuBIgAAAACghUARAAAAANBCoAgAAAAAaCFQBAAAAAC0ECgCAAAAAFoIFAEAAAAALQSKAAAAAIAWAkUAAAAAQAuBIgAAAACgZe1gqvnc67vf/W7zCQAAAACwaIuMyUYHigAAAACAi4FbTwEAAAAALQSKAAAAAIAWAkUAAAAAQAuBIgAAAACg5VQCxcuXL0/W1taq6cmTJ81ax6O87t2713xbLWqD2CarWo/T9PDhw9I2L168aOYszu3bt1vHR+mk+u0s1C+0bwAAAGDRFn5FcXNz89iBkQftb968KdNVd/Xq1eYT7PPPPy/Ta9eulelZu3LlSvNpcR49ejS5dOlS8w0AAABYnFMJFBXAbWxslM/7+/sT/QUOpd3d3TLvwYMHZ3KlaFncunWrtMfW1lb5fuPGjTLFO+pDZxUkKUBzn1UZlM4iYH379m25Og8AAAAs2kKvKCpA2tnZKZ+/973vlanVblfV7YeZ5uuqpDx+/Li1/vXr18t8y7d4Oo3hdU/qtlCVzXl68P/q1atqEOJbDvP6NTFfJW2rWyc1zfmoPXObKO88z7dealm8DbPWFr5FNKZ4m2Ysnz7nfcV1xf1AQZJSXFfpOL9g0La1/Lry1bxasJbzcb9z2Z1y3aS2jqaRl8nz58+Pvsf5UW5TpXzu5NtpfSzj/CxvozxV/lq9AAAAcM4cnJKNjQ39If+D/f39Zs47mj8NkJpvBwfT4LHMqyUti2rrOGmfkctQS312d3eP1quVfxYxr5hUNrVBLLP2VVvXKZalK18ntdv29nZrnrbP22n/fe2fU1Rb7uTjpjp6Xvwc50V5eU7z2traqubnVOO2UjtG+Th5ea7f2OMV22CoD+T2qrWpUyx3Phdqx1xtZHlZTLFeAAAAOJ+WIlCs8UA2D9LFg+44sB2rr1wnLQ76VWaLQUOsn+c5yDKX2et25SteN1Jb5/Z2GWp5igMr5Sc5D33W8lxW5+vtRPtw3rHdnUfm9XPd5uX8chu4zl39aKgcXctzW0ntuLiNa/v3stheWe4X5nrlMoi3cRKv72NZq5fPx1qeAAAAOH/eH6WfEA9IawPd2oDTg9OcaoNoByNdA/xI+8l5Kp1UENKnb7DvZS5H/h55IO/69uWb5W3N7e3gwG3qoMPtJjkPBw19KR7frr6geVqWef2T0rWfIX19WGrldFvl/fmY5dRF7ZfPkcjHayhlPq65P0Rdx3cR5wwAAACWw8Lfeurnm+JzX3rGSy+4OUl+jkzPmJ2Vp0+fdr4IRc8nit+m6XX1HGe2t7dXpjdv3izTvnwzb5ufs/MbY/0iHeUpfgOr2m0aTJTPr1+/LlPn4XX7xP15X7G87geuU3SSL7Lp288Ql6OrnfX8YC6n2yrvTy/IcXtG6qO1Z/5qz0ZGY45BLpufuZwGsaU8Xe7cuXP0LHF0Em8sBgAAwGo4kz+PId/61rfKVINkDbg1iJ4Grkdpt3lD6iw0sPbg2i/LUT4x39pg/TRpYJ4DAQXGDjIchKjcWlcvEIk0ML97925ZVwN4q+UrsQ3EAWkOXHKwGQNH5+t8HJQ4D8/f3t5utW1Mz549K+vE4CTKeZrXz/P9Uhwtn0ctMFI9lWct+HE5YltGLketXSXO90thJLaRgzG3vbn9c94qi/NxQJ/Pm5hyvl1BbOSXE33yySetvPb398vynCcAAADOqekg8MRNA5DqrWtO08Fts+a7W/WGUlZbR0n7lq5bWWOK5bB4S99xb7XrKoNvWXRSOYZu54yG6hbrpc+1dbqSyuH8ffuky+t8xxwztV1t3zo+tfpbbP+cfGxnNdQffZxzuWrJ6/aV08n1qi2LKevrCz4mUlsek/KRvnp5HRtqA9cfAAAA59vCryhOB6atqzu6ouWrFdF0QNp8qqtto7x9xeP+/fvlilc0HbgPXlH0FaFpcFG9DXQWKoPKFOm7r6aarhTpamGtzlp/epyab4dqdTO1S2zffCVN9c/5ZW4DX7XS1U/x1TUds748VIauttNVOudXo+1qdVOe817N0nY6ntk0KCr1mOc4a5vcl5RXnucrj7X+qnrW2lF9odZPta6v1Iq+1+ol6ku+Aj1Lu2ld9Tm1TXSctgIAAMDqWZsO/vqjBgAAAADAhbLwK4oAAAAAgOVGoAgAAAAAaCFQxErwG0rHpq63lQIAAAAYRqCIlTDri2wIFAEAAID58TIbAAAAAEALVxQBAAAAAC0EigAAAACAFgJFAAAAAEALgSIAAAAAoIVAEQAAAADQQqAIAAAAAGghUAQAAAAAtBAoAgAAAABaCBQBAAAAAC0EigAAAACAFgJFAAAAAEALgSIAAAAAoIVAEQAAAADQciqB4vXr1ydra2vVdPny5Wat5aJyxXLevn27WVKn5ctal0VTOwy1l7148eKojZfdLPU6SU+ePGn1RaWzKMesXO6x54XqNKae/v/k4cOHzZzFcZ00HePevXud9QAAAFglC7+i+Pbt2zKQWnVPnz4tdbnoNIBWO4wNDvb29sp0Y2OjTJfVrPU6bctSjj46J+TmzZtl2kf1efz4cfOt3/Pnz8v0xo0bZbpIjx49KtMrV66U6ZBZ2gAAAGCZnUqg+OzZs8nu7m75vLW1NTk4ODhKDhD0m/dl8ubNm1I+lxvjqN3k6tWrZTrkzp07pZ3VR5bZrPU6Sbdu3To6X7a3t8u8VQg87t+/X8rs4KqLg3DVzfV0qm3rZdeuXWvmLM6rV6/KdOy+1a9VVvVzAACAVbbwK4off/xx86lNt5XpSmNMuk2xpraukm/3irc3KsXbxvK6s/Ltcr6a6Pycusosvi1N6aToykzcv+ralX9cTym2QW5T1SPmndvL69+9e7d819Tr5vVdpq7lkvflz7HNuq6qebnTouo1j3xbtuqnPLt+ceIrVF1XtHJfd+rqh7k9tO+u/Wt+XFfbqu3i+ZTzU+qqi5dvbm6W7w8ePDiaF/OUeIyc+tT6mNsmiv3JScck83Hy1cy8TSxvbqe8PKsds7h+LqPE713tCwAAcJIWGihqgORBYrxSo0GZB+bR+vr6e4Mifa+tK/7t/7LyoF9XWY/LA+N8+6va99KlS823Qx7IZrr1z4Pk3HYffvhhK2+tGwezQ23dFdRZvkKW92UKJkzrKDCxZaxXFx8vBx6m+qkMXVcudWVTx7PritZHH33UfGpTPbPaeaZ91/avsmp+pG3VdkO3YZ7G1c++W5VVL/+/Eun/j3gu6P+f2J9MxyT/P+Mryl2G2kBXhWtUVpUrU/ldhrzv3OdUh65fBAAAAJyYg1Oyu7t7oOxraTroa9Y6ONjZ2Xlvnk0HeWXZ/v5+M+fgYBpkvTdPNE/LIq+rspjLlde1oeXmsp0VtVfev+sby769vV3mjWlfTfVdSduZ89Cxyrry7uJy5+MnXqb9xP6jdV0GH8tlq1efuP/YF8V1rvF2Q30xU/2VIrdnrH8sl9tKcluLz9Ocb+R2H1IrXxfvN5Y7cp/Px6pW35q+9bzvWv+ocXt29Zuustba1uXK+3ce8dgAAACchoUHinmQ5IFPX4qDuK5840DXaoMqb69lNUPLbeyg+LR0tVtWWycnt0/XoNz7ym3sthoajEd9QUJcpjyVtwfJOaDS56G0yHr16drPkK5yR14np3yeOYjJqZZ3V55uzy5apytIMpdjaD1zP6jt23kNnauR+1FO7mfRrMetr994WVdZc/92vXO58noAAACn5dRvPZ0OjDSqKWkaBJTbvHxb4Dziiz6Ut+l2ruPku2r00o9Yf9Mtg76VMt5S2ce30fm2y3wbom6Z1bHLtz/O+sIX3S6nWxdrtyZ6mW+zc95+KYhvwZRlq9eQrv0Mcbm7buVUW3Xdhp230b6nAU/z7R3dxpjPG7X5NEBpvr0Tb4/MfEyGbjv1W2/H3p7q27Vrt3LOklfXrb9We6Oq23/scXNZa+XxMt0uXaNyTYPA5tu7PphvO83rAQAAnJaFPqPowb4GO37WzIOqGFDmpLcpmgZ7HtgqWPI6Gjwp36Fnd2rPMp0UDaJjoHZa4rN5sZ08uPcg04GS2jyul5MHwh7MxkFzX3BXW98D8tqAuG9g//r16zL1MuXtwDCXYdnqNYb2VesXCgRyMGAuR+15OPU15bm7u9uq83blLamqq8q+vr5egsW4vtrQ/UX8cppPPvmktZ6DzLhu5LLW2jMaCn6zMYHR559/3nx6x/Xw/zMKfmv9RfOUasFgbd8+92rH0m1QC2r9S4f8/4+PjcQXfdXy8j7Hth0AAMCxTAdLJ246uCq3R8XkW6h8u5jSdIBW5tXWj8m3csVta8n5mW/f6krar+h2sNpyp5yv9G1Tu03OdfQ+jyPvL6doqG1VD98WV1uW50V9eft4q+1qy5XcFjEffY7HWZ+Vl7/bWddrFkP9UOWRWM+u5HVrdcjJ6w7t3+eXDLVr7Nu15U7et9XWcdIxtqGyxnWltk5MXr+2LKdYt9gHa0m6+pdT7iu1dZy8blcfyMe79n8SAADASVr4n8fQb+6ng6Hm2yH97bE8z6YD16Mrin23gE0HUu9d7dB2mh9NB3fNp+PR1UyVLdL3aZtWryjo6oR861vfKtN56QqETAeyZRpNB9ll/5HaNreBaf2+v3mX33qZ1f4WovalMviWUV31GpKPm688io65r65MB8dlKmddr1moH6pMNTqOfWXtom1ie7h8cZ6pfbW8VgZtE6/Ya12di0N92/1wjFnW7bpi2UVlqnF7+P+MfN7re9//BbX/p0TbeJ/zlDUfH7frUL8a6rMAAAAnbW06SKmPtAAAAAAAF9LCrygCAAAAAJYbgSIAAAAAoIVAEQAAAADQQqAIAAAAAGghUAQAAAAAtBAoAgAAAABaCBQBAAAAAC0EigAAAACAFgJFAAAAAEALgSIAAAAAoIVAEQAAAADQQqAIAAAAAGghUAQAAAAAtBAoAgAAAABaCBQBAAAAAC0EigAAAACAFgJFAAAAAEALgSIAAAAAoIVAEQAAAADQsvBA8fLly5O1tbVWevjwYbN0+Vy/fr1VVrxPx/T27dvNt8V58uRJ69gonUU5Xrx4Ufa9zP0483moNjxLnF/DOL9W7/yaleoW2/k81xUAsDoWFij6h/3bt2+bOavn0qVLzSeYBpM6phrMLoOzKMcXX3xRpsvSBkN0Lvo8vHXrVpkuA86v93F+rd75dRIuUl0BAMtrYYGif9jv7u5ODg4OWunOnTtl2TJ69uzZZH9/v3y+efNmmeKdN2/elOnVq1fLdJEU5LgPbW9vl3lncYzcBleuXCnTZXft2rWjdjtrnF/9OL9W7/yah34Gqp03NjbK92X6BQ4A4OI61UAx3rr04MGDMm9zc/NoXk2+Bae2nn7bGpf79jnP1+1s88q3W6k8r1+/LsvyIMlXSWNyWbJcL5VV62qa83H5a/WM83QbmD/fu3fv6LPWqfFyp3gbWS6fypT3FXn9u3fvlu+aet3a+rPKtySqfspT05qnT5+WaR5M5nxUr9hWSrWy5vbQOnn/+q5ljx8/Lt/X19db2yiPLB/T3Fdz//OxjPNr+Y4R81Wq9ZPYXvqcy6Pvx5HzU11O4vzK+Sp5+1nOr1h/H1+l3Gdq4nKl2K9yfzrr8yvWzfn5HKtZlfMrlyf38dxPxpxfuf9oG63jfjSvePydl4JiB4uR2yKv38XrxfYEAGBmB6dod3dXlyw6U1Zbx0l5WV62s7NTnTer6Q/o9/JR2traKtNYhq51lba3t5u1Dl26dKm6npLy2d/fb83z9nm7vF5fivV3+WtJ+xftM86vlTnWP6+fU26DsYb6TNdxVXmVsnicvDyXPZe179jG/fetpzS2vWK5c1/WPmrHbx45D+WbxeMeP8d58+pqr+OeX33rqryznF+1OtdSLMMqnV9D/4d05avyKmWx7b08lz3n2Xe85j2/8nmTk816fvUdh9r5M0ZXWV2OmO/Q/4c1cRsdbwAA5jXfiHNGHpz0/WD1wCkPKmo/PM0/DJU8IPMP4Vl/QHo/zsfiYMW61vW+44DK2+fye11NzYOSOACSPEhznto2DwpyHv6eyypuc7eVj5NSPA7OI5bVuvKeR9x/bgPXucbb1fqIdC2vtbfbM9Y/lkufo9r6Wa1fmOuV6+ttnNz2Wv+47d13PMXLlWJ93V/mcVrnV9e6tePieuW2Vn61PLV+PPbazmVw2znP2jFxe7kNY161ctWOR1fe8/L+875c59y/xeXWOjVdy2vtXTsusV3y/mvrZ15HKXO9cn19HJ28XG0d21vLcvv7uNaO15BY1ljXWB7nG9sltqEcpwwAAIw136hvRv4h2PXDvm+gJFqWf1jHH6LH/WHpvGoDIS/z/v2DvmvQpPW0XMYMcqK4reX9SxzY5rbLeejzUPIgpOs4ebClskSz1m9I136GDPUvL8/9xG0V9+f2zqkrb7d/HshFHtT1pVw256t00obaudYuonmxH47lNj3p86sv35q4reX9S1wv963cdvo8lJbl/BrTV2u6ym1envuw2zHWy+2dU1feY8rs/fSlnL/zVepTO3djX5lV7FtZbq9a+9lJ9w0AAGoW8jIbP9/S9SIEL6+91MbPJOVt9/b2ynT6g/LYL8NxXrXy+fkpv0zCZa09dyPPnz8/esZkqN6Ztp0OTJpvh/LzW3peJr4F0S96cBvou/Nw2w3xc0evXr0q0/ziDNVDeeolKNFJv2ijaz9DXO6h/pWfV3Jbxf3p83Rg1nx7R8/Y1p4Lcht0vWjDx2tILpvLXCvLcQ21s+sUl3edh2Oc1vnVl2/NmPNL4nruW9/85jfL1N/VNqt4fsmsL0pZ5vNLdLyG1Npchs4v1yHS/vT83zxi/83yeed1c78Ql3+e8xEAgLEWEih6oNE1QPEP8TxQ0AsD9PIbuX//fpna0OBlHp9//nnz6ZAGgt6/9+Oyer75hQfy8ccfl6k9evSo+fSOX/jgwaa2l1yfPCDIA1st90DGQYmXeXCl5QeHV4+ryQMR7+vGjRtlKjnPqLa+6qN6dQ30h2hftQG4BqF5IGouR9dgstb/avXyMVxfXy8DyNhGakMP5CK3fxzM+diq/8b5Mb+c8rlRG2SfhL7jKV6eB7Nu467txjit8yvnK2p7HwPR9pLLn+uV1/NyH4c40F/F80vcJpH+73X7Zi7HMp5fouWS84wp/zJxzPml/2+0H/W3mNfW1lZZ7r4yK/WhvK32lc87lU/r5hfS6NgrsNa6swb9AADMZPqD71T41phamv4AbNY61Leu0nQA0Kw5/SldWe4U15tVLT+l6Q/j1nftI37PKd5+NbSukg21QU4qV8xfn30LmJLl8uek29669u1b4mKK+vLOt6GNEW8HqyWVR2I9u5LXFfW32joxydD+a7d51dZzcn+stWNMakfpq5fXmcfQ/q22ntouH+fYtmPF7WPKec9yfkltnZh8DGY9v/x/lL9r+5iH8+07B5TUVstyfg2dN+5jq3Z+DeXrYznL+TXUX/L6Y9WOufOL31Wnvnq5Tlkstz4DAHAc7ZHJCer7oVz7IVcbIOaBw9Ag0gOHecUBj8sY5ylZnt81cOgqc21QlAcRtXlO2l9s47hubt+uPFyGoUFRTFlern0dR9fgKB7bWQeyuX7aPs8TbaNUK0OXWtsp/6yrXrGcXesozTswlTwIzclUltrynGKZZ3Ea55fE9ZxqZcz1q81z0v7jsVX/i8cn6spD68synV9d547KaKt4fnWVOfabWc4v5Zf7gFNsq3nkcii/PM91rNWrb//Ox+cXAADHsaZ/pj9YAAAAAAAoFvKMIgAAAABgdRAoAgAAAABaCBRx4vy2wLGp9pZTAHWcXwAAYBEIFHHi9Jr3WfT9jTQAbZxfAABgEXiZDQAAAACghSuKAAAAAIAWAkUAAAAAQAuBIgAAAACghUARAAAAANBCoAgAAAAAaCFQBAAAAAC0ECgCAAAAAFoIFAEAAAAALQSKAAAAAIAWAkUAAAAAQAuBIgAAAACghUARAAAAANBCoAgAAAAAaCFQPCfu3bs3WVtba6Xr1683S5fDixcvSrkePnzYzFk8tclptIvrpnTWXA6lZesDy+As+2HtPD2Lcjx58qTsW9NVorY667YDAOCiIFA8BxQMPHjwoPm2vL744osyvXz5cpkumgKE58+fT65evdrMOTl7e3tlurGxUabL4jTquurOuh9mZ1GOR48elemVK1fKdFUtyzEEAOA8IlBccQ5+tra2JgcHB6307NmzZq3l8ObNmzI9q8Hp69evy/Q0Bpd37txZmjZXOXZ2dsrnmzdvlineOct+eP/+/aPz079UuHXrVpku0qtXr8r02rVrZboqfJ6dZdsBAHBRECiuKAU7uvVqfX29fH/8+PHg7Vjexkm3wUX5trjbt2+/N1+B6ayUj7ZVGUVldn5KtfLmsuZbKJ2nk8sa53kb32a3ublZvuvqa229WTnfmFyOKC5XW+Z2Pq6cn46RA4EbN26UqdXK3HVMc75qJx0rTXM+tfZXUt7xezxuMf/aMaiVNfaVWjvG71oeed+z9MNZ5D7rturKV7/guXTpUvPtnZyPxHZTynWTvI72q7zUjqbyaJn2LXF9pbiu5XV8rC0fBy+P8+f5f8Nie7ifKNivXb3PbVDrV5HXq7UnAAAX3gFW0nSAeaDDV0s7OzvNWoe2t7er6ylNB1vNWgcHW1tbrWXaLu8nrj+Wtol55LS7u9us2V9WlcVynqrz/v5+a57LqmVxfk6q9zxU7pxXbvtcptpxm3f/kvNy8rGMZukztXWclHeuu7eP85Ry/ftS7Ad9ZVUfkdxfa9to/zZLP5zFUP+q5et2qR37WA/34VzXWY5XbIO+dlWK6+Z9xuRySV7vpP7f6GpX7y+2Xe1cjKkmbhPrDQAADhEorjgPfrvEwVbkgWptkJQHfl6uzx6kz8qDsr7tXVYNMjPXMw+6NU9Jyz04rW0vmt+17Lhcvq4Bp5fH/fsYaNk8XN/cppqX8/X+4+BatG1etytfrxuPgftKrndXnto298mch8ua9+/2im0YB/sxeHKeub94/Zz3vOL+cxuonLGskdsgljnqKmetvWvHpe9cGtq3OM94DM3HMtfXZXPycn2etb272tVlV3L5tdzzYhuIy9pXVwAAUEeguOI0COoajErf8toAUzxQrw0S59W1r8iDur5UG/DF5V0DUg8mNZg9DSr7rMfBg+FZB9HiAXOtPbzM+ebvWSx7X741tXq7XrGt9d39KQdxMQ9vO5TM/SqX13048/p9/XAWXfsZktsg6yqn2invz8csp668ve8c6EU5r1rK+Z/k/xt97eplLn/+HqmMWtbV9wEAQDeeUVxhfp6o64UlXp6fKbKnT5+WaX4hhJ+dOskXswy9QETPML19+7b51k3PK0W5bt5PdpovsnHZu46Dn8/K+3b7z/PCGW+bn0EUP5/oN5563W9+85tlGuWy9+WbedtcLx8D5+l+6PKofOpf7ncxD++/j7Y17yuXQX14GkA0394Z6oez6trPEB+jrpexuB1yOdVWeX96wcs0WGy+vaNncmvP3g29yKb2nGJNLttJ/r/R164+hi6/163Vx+04zzkGAMBFR6C4wvKAPPNATi9vyYM/Daw1wNre3m7mHHJQc9IDKw3YNIiMgzm/7EIv3YjzDw7evbk1pziw1rZ6MYkGifv7+2WevtcC49qAUXUd88KLIf7TGF1t1rXcZTrOmxv9px5Mbeo/leJgz0HUhx9+WKamPrHevAzpJ3/yJ8vUcr7iF4W4jzj47qqX95/7qfqdy+R+6WUOJrcqb/F1cn5Sa8OcZzTUD+eh+rhNor6+pW1iwJupjrmctXqpzNrPJ5980mojnw+xrcyBVeRjq334/w3tP+aZUyzbafy/UWtX9ZscLKucWjcHxaqTzgWte5xzDACAC2v6Ax8rZjowKrdT1VLX7WC1FG8N9O1otXQSt23V8nXyLWN9ZVBSXSTXX9srxXlK8XbEk65f3zFwOaW2Xq0sY2/1NN9Sl1M+3tp/rW1iin2mK18n5Wddtzx2JdU75i++xTLm6+VdSfvt2ndu25ivxGU5uR/OYqjPum8NracU+0BteUyuV9/5rZT/PxjqCzaUr+ojffU6zv8bXfnmcmkf7kO1lI+/xX6Y2wgAABziiuIK0m/Ux9JtYNNBV/PtnemxP/qj2+Lb0U7LdDDWfHpnOjBuXZlQeaaDvvI5Ux18S1uuv65s1a6AxdsRlfd00Nh8OzQddJb962/bzWqWY5D5zzMch66QTAf9zbdDaqPPPvus+XZIbaD2VT0zra/58WqLPteOleh4xStUuuVRbWhqX+UX52W+Auhj4e/xWCmPfKxMZdN+uwy17Zh+OAv1q9r5Jcpznr4lKlOkcud5ouOh+bnN3bfzlTTVsZaP8tf61vX/hugc9f8dp/X/Ru3/ApUxX7HUFWi1cVedaldUJfZDrjYCAFC3Nh0cvD+CBAAAAABcWFxRBAAAAAC0ECgCAAAAAFoIFAFceHrj5yyp9pZTAACA84RAEcCFNk/QN8+LbwAAAFYJL7MBAAAAALRwRREAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWggUAQAAAAAtBIoAAAAAgBYCRQAAAABAC4EiAAAAAKCFQBEAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWggUAQAAAAAtBIoAAAAAgBYCRSC5ffv2ZG1trZWePHnSLF2se/fulf2rTH2uX79eEoDxdG7pHBvjxYsXR/8fLLtZ6nVW/H9bTA8fPmyWnj/+uaJ+lF2+fHnw//jT5GMxRMdH653Vz0MAi0egCIxw5cqV5tNiPX36tExv3rxZpjUaeDx//nxy9erVZg7OEw3KPJB20i8FNNUALwYwOUXexikP9vLAvTag7VIb9Cv1DX496MzrL+oXHg5Kxp43e3t7ZbqxsVGmy2pMvRy05KTjeJYUMJ1HatfHjx9Ptra2JteuXWvmHtJ5+Pbt2zOt+6NHjyaXLl1qvnV79epVmWp9ABcDgSKQ6IfgwcFBSfrhqZR/uC/Ks2fPSjnu3LnTzHnf69evy/S8DrJW0VdffdUagOs4zkPHdHNzs/n2jn4xIENBTl+w519CiNZ78OBB8+3QafZ51evu3bvNt3c0mF7ULzw86B173ugc1Lk477FclFnrFb1586b5dPru379/9P+sg+9bt26V6XmiwF3nlupYC7Dc5mf5i76xgeo3v/nNMtV5CuCCmP4nDayUaeB2oK7rlOdNfyCXedn+/n5rO6Xd3d1m6fu8fld+sr29/V6eef3afvN85WNbW1utdZVq5dS8vF5MLkfXfMnlVzvOQ3nGfFw21THLx29nZ6dsr2k05njltnI7xvlZrX1P0gcffHCU78uXL5u5s3M79fWn3L6e7+OR2zTWXZ+t7/yp9fHacRXn37VcnF/s8+L5+RjX+kHcVnWMy7R+rE+sp9TqE1PMu3aO5fzyvvw57kfr1Hi5U8z7NOtl3sb9JZczb5/3URPL6JT7YaZ18r5rxz3WOba1kvttnJdTrRy5vLn/5z7gcsb5ffXzOlk+vjnl4yu5zkpRPj6uS6xjbOe4bi3VOK++c9xieQCsJs5erJz8Q8+fY4o/DMUDoVrSD7MaDwS6ltcGMk5xm9pAI/9AV/IP3qHBgI0ZaNTKGAdCeV+53Yb0tYFSNFRetZONPV55vdo+PODKxyGnMQOfPl9++eVRXpubm83c+bh/dPU91TsfKx9LbeN2yNtrGyUt8/bel5O3GTq28XiZ8+4Tyzmkrx/4uObyuwwxxbLmPp+T2s5qfSYul7y8K8Xt+srg8/M06yXx+Eo+dvn41/bvY2C5zDF1nV/eT86r1vZxndw39L22TUyxvfrKqrqazyUn7afW1jXeR2576du/kpbbUL3ctrlcziPOc93y8c0ptkHkdo9t2cXr5mMLYHX0/0QHllT8wRl/oIrn+4ezf3jqh1bkAUDXD0T/IB/zA9FcrlwmyT/svY7KVyuDf5DncmfatqsO5rrkcnXNH8ttGweBzjOWO9Y9riu5/PMcLw9InMTtp+3ioCjv/7htIAoMnf/e3l4zd35jjmnk9nWb5e+mee5v+pz7pJLPG3/XOpGPg/LJND/vM/P2OeXt3A+69qNk8fjG4+hj6zqZ16/l3cV9LPcf8TLtJ/d1l8Ht6O+1dvJx8T5Os14us8uVv5vnx/7o/dTqkPk4uk6Z+0Oui3n7vLw23+3ifXmdOE+8z9o51tUO3sbJ+9X6Xe3gdftoeV87uq2Vchu6vrFviPuRpq6PUs3Q8cm62gfA+USgiJXUNVAS/0DXVD/M9Fk/DGv8Q6/Gy/p+gMYfwjH1lUtpzA9llz0PAiIPIrrqZ85L5bXavFn5OOSU69fXztG8x8sDo67tPBjqS0NtWKOgsJZXXxoKJMce0yjvIyaL/Sm3h757nvbfdVxjyv2mq796sJrlMii5zm6DoeR+5nMr7zvWKYr/R4yleijVxGVuO+ed+6w+DyW1pZxWvfqOb96X5uV6dx3rvuPWxXVxnTPXyX1DavMylVnr1I6Zl/Wl3IaxzcZwG/WVsasdI7dPX6rtI9ZxqJ1qbdTFeQK4GHiZDVaSXwBQewA/vszBL+zoevuiXgoyHcg139q0j+kP0OpLPfwmSr9UJLtx40bz6R2XZTo4GPWikDFvPB37Ihu9JEJ1UXn9ghO/IOezzz4r03nohRTTQU7z7Z319fWjty9KXztH8xwv1UcvY9D8rrfxxRe3dJnn5R96Q+enn37afJuUz9P/V3vT0Fs9Z305Ud+bRcXHO/anmLfbTcvd38e0V375hreJ831szG/blPjSKCXxues3jA7xeeTtamWqncOzvvDF9aidi17mvPx/k88v/z8iY/+sgN+yfBr1Unnzi4sil1/cd3J+PtaxPXRsdd7X9J37LnPXi2zcjvEFKn4RUu18V5nVx3RM9H9TrI/4eA3pqvM0GC7TIV6/71i4bH0vsnE+ffI+9JbVWEe3cU3su0Pcf8f8Xw7gnJj+gAZWznSQVH6rqWnk375Of5CV7/7Nc+7q8Tfftd9ke7nzybTfvG/pmi/KK5ejz5j1/VvuWIeusrtt1Cbx87y877yf2m/J9V1JZctiHvMcr9r+Mrdl7VifhHjrqZ5VPI5Y19rxUf/SMq3nuudjIPkY+3vczn3V+3Q+8/SPmL/5ePrY6HP8bu5LykP66lXj4xv37To5z8jrR7mskZfV2iP3v/h/QC6Dv3f9H5GdRr28Tj4XamXrapO8n9wG1jU/yvusiWXuKr9oP1oWl7tesQxeZxbxuI7hsvTVvXbOuM18fPvqW6Myan2XVZ+VlE/mfeUyOo/M5a2dBwDOp9n+pwSWQBwoaZpT/mFeW8cp/sDzD+S+5B/WtWU5aV3/IK6lXM6+dZVqP5y72kAp//DP+dcGDrPwYKIrxcFPXzmVaoO4Whp7vHJb6XttvZiO66TedipD7aW61+qvNvf50ZcyH0v3yaG+qOTjW1uWk/IbKlc+H4b6l/pMVzlr7Rf15a08h9bxuROPgT7HOupz7HdWO24xqeynUS+1V56n/GSovznV1lMdx5xfPr5j9jV0/rrc0VC+Mc+hdX18++rldbp421pZra8vuLx9ZXCSnJf7cZynFP+vnaV+7pND9Tbnk38OAVgt748YgCXnH1j6IZcHVP7hmuUBVO2HXd8PbSf/8M371ffavL4fxB442dCA4Dh1s7jecSkP1THvv2tg1DU4qxlTp7xOTLW2ysfHKR+HeSk4jPkeN1jsKq8DtFr9taxrO6fcljG4iW0R5+dkfevEJC5XbZuuvt11Tnr9obrGFNXKkAe0eXlMbsN4DDQvnsPiPp/7WNe54DKcRr1i2Zx8rna185jkYxHr5PoOzetKuT/EeuW2tKE6qE0jtUttPbeJdK2j5D7Qxcewq7yW8437t67+EPPW57hMbVg75rlta8ejxst0LIY4zzHrAlhua/pnekIDK0PPX+gZm+kPz3P5B5pPk54j0rM+08FC7x/xB4Blomd79YzyKv2/r2f/9AzgNGAa9Vz6snLbj/m5oecYNzc3+fkMnBO8zAYrwy+Q8YsY9MNI34deDnKR+eUhTn4hhF4GoYAbAJaR/l+P/3cpUBH9vx9flLXM/KIpv6BqFenn7tggUbQOQSJwfhAoYmXkt9dZ3xvjLrq+t93RbgCWVdf/9zL2LZ1nTW+F1k1bqxw0qeyqw9g7UHTcCBKB84NbT4ELzrdHjcVviwEAAM4/rigCF9wsQaL4b7wBAADg/OKKIgAAAACghSuKAAAAAIAWAkUAAAAAQAuBIgAAAACghUARAAAAANBCoAgAAAAAaCFQBAAAAAC0ECgCAAAAAFoIFAEAAAAALQSKAAAAAIAWAkUAAAAAQAuBIgAAAACghUARAAAAANBCoAgAAAAAaCFQBAAAAAC0ECjiXLh3795kbW1tcvv27WZOv8uXL5f1Y3r48GGz9Gw9efLkvbKNrddJe/HixVEZ+rjMmuL8ieeLPp8E5dPVr3Uuen9Ky3JuAgBwkRAo4lx4+vRpmd68ebNMuzjwefv2bTNnNZzU4HxWe3t7ZbqxsVGmXR49elSmV65cKVMszqy/JJlF7XzR5+vXrzff5qNfKCifrn796tWr5tOhGzduNJ8AAMCiECjiXHj27Nnk4OBgcufOnWZO3RdffFGmu7u7Zf2YhrZdlFu3bh2VaXt7u8wbCoBPi9pE5VD79vHA/tq1a2WKxXnz5k2ZPn78+MSvvn344YdlurOzc9Qn5fnz52U6L5f56tWrZZrdv3+/7OvSpUvlO/0KAIDFI1DEytIVFA2MY6rd+hhv5Xzw4EGZt7m5eTRvVr6C46SBeZ6nKzE1uhIT1xu6CuQrpV1X6uKtoTHl/ee2UnnzfIvt5dRVTtfHgUPeTnnVjpPLVyt/7RgOybcqap9KrmeUj5XqoO1rV8ly2fM6ua18hSzOz4FbrX2PE9zpam4Mqu7evVvynKcdI7WTrvopSIy/RNna2irTrj7ex8dJZRSX1Sn2M+Wv/ateuZ901S2fX/GKZc5DZYnH96yu2gMAsLSmAwxgJU0HrLq80Uo1u7u7760X06y2t7db208Hsq3vTtqvTQfb1XWUtH0XLetbvrGxMSrPvF6tPGpPqbWX1q/pqrvT/v5+tYyaL7V9xXYbo6sNlHK5a+s4uf4yts/kdlRZ+vplX3upXx2X2jXnOy+3a9Y1f4x87uQU28DHoKvNor7zS0nyMa3lexLHAACA82L+UQSwJDw41gC2j9eLAcFxeKCpaeRAweWJA1QHSOb5teBo3vKqPLlM5kG+k3g/tWDQ6+dyRx6k17Y37yOXq2v+WG7bOMB3nkqx3NpHXlccvPgYxO3zcXEeua45UPFytZ/7gdsy7/+4bVAT+9y8eWs7lz3qmj+LMXn4uCjF4+BjYLGumc/FeLy8fdy/8zip/xsAADgPuPUUK+/169dlOvQcn1/MchK3mPm2uOlg8+h5K/OLXcy37U3Pt/eetZoOYMs05yFjyutb+WLqe0mI9zMdEJfyiMqkz7VnNLX+dGDd+4yYb4/te+GItlc+Klu8ZdHPwP3Mz/xMmc7Kt+TqlmLXf319vTzbGdtb7eTbKPX8W+Tn4fRsqHz00UdlOg3gjuaZbxHNL1uJ32Nb6tlOJd0q6dtzY1ldXlH5ToJup9St1aby1PpXH9/amc+prvmzGJuHy5yPg9pJfck++eST5tP7tz7ruU3x8YnnbXzu1vs6q2eBAQBYRgSKWHlj33g6dr0xHJzWXsbhQEjLPDBVYFbTF2R5cNtVXgWDftYrq20TB8k5mK3x+kPtNfZFNn7+zC8U8jNwCuqGtu2i7RRIZArG4vOEY4JZU0CnNqqVqasPeX6tLOLlfWLwMw8/e+ngSGVRkDgPB065f7vfHOcc6so7U5vlX1LUgswxL9bxvrrO21n6BwAAF8Z0IAGsNN1CNqYrj11vjK7b4uJtcNOBepnn7/H2N3F5um5383LnE3n/cd/SNV9cNq0zhsqr9XO5M62jska+5a/WNlo3fp6X2kV5KOU2mgYYrdstXZ5a3b3MeWi72rper1bmvL/Mbdl1rI/DeTvVjv2sasfH9T9uHXJ7i/cX8877F/fv2Cd9vGrnSVbbXoaOHwAAFxGBIlaOB5VdKQ4C+9Y9zsBQA1htr2kt7zhY98C4luJAOA/4a8kD6b48nbxuVxmVagPm2npKedAuMVirpSznP2Zw38WD/q4UA72hPhP7Ql++cb2+41Vrq9p6MeVjMSS3/azbD4l5O8X6z2uoP9baX3XNfcdtPNQPXObastp5FM9dAAAuMm49xcqZ5XmrWdadhfLVrZ96zmk6EG3mHv5h+ul51XqmSrfrTQezzbd3poPfwb9P2EV5xv1OB7xlv3GezdIGuhV0FrotcDq4b769Mx1sl/Jk8ZZBbTfvLaeieqnetbbVvuOziDoeKlONyhHbSNt11Smul59THNJ1fER5154R7eNnWFV/5T3r9kPUPyO19Sx9qUutz7v/dtVB/ST3Td8+2nW8ROdjX5l9my4AAHjf2vSH8/ujOQCd9Oze+vp6GdyOedYP7+gZM71oRQP4eYNkAAAAnD4CRWAGuopYu+qmqy/HuTp2XjkwrNHVtZO4QgUAAICTx62nwAy6bs0kSKzrCwT7/uwHAAAAzhaBIjADXYCvJdTpmbNaeyn51lP/WYexyX9mAwAAAKeHQBHAmZr19lOuRAIAAJw+nlEEAAAAALRwRREAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWggUAQAAAAAtBIoAAAAAgBYCRQAAAABAC4EiAAAAAKCFQBEAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWggUAQAAAAAtBIoAju3JkyeTtbW1Mh3j3r17Zf3bt283c5bTrPXCZPLixYvSZkp9Hj58WNbR+mOor2h99R0AAHD6CBQBHNujR4/K9MqVK2U65OnTp2V68+bNMl1Ws9YLk8ne3l6ZbmxslGmXzz//vEyvXbtWpkNWpc8AAHBerB1MNZ8BYC7Xr1+fPH/+fHLe/jtZRL2++uqryTe+8Y3m22Ggpf2ed5cvXy7TN2/elCkAAFguXFEEVpxvj3RSkKHb82q3dfp2v5jibZXa1vP1Oeedb8H0+gqmJK6b1/etg13L8768vgKKuG3t1sOzrNdxfO1rXzsKEl++fFkCUu17rFgmB15xXq0PSLw91KmrTr5N2EnfVUa1ueX2VOrat8qp5W/fvi0pbxdvRY3HrbZcnF9tna75kuuVl2cuyyzHBwCAlaYrigBW09bWli51VdP29naz1qHaOk47OztlnUuXLh3Ni5/jvKi2Tkz7+/vNmvWyRipDXt6Votpyp0XUax5ffvnlUV6bm5vN3NnFMnWVeWNjo1n7kL7X1lPKfaavHXZ3d5u1DsrnvNxtn+X1copyWfNxkrjcycenNl8pz48p1ivy8q56AQBw3hAoAivKA14FYFFtwOsBfx7keoAfgwkFC84jBkTOI3OAN3YA7XLnAEbiIF4cXHpdlSEGC8tUr7EUGLoce3t7zdz5uZy5rLEt3Rdye5rziG1bq7/bVamL8tby2MaZj0Xso32839zXzctj+UXzlCLPy/t2fbv2AQDARdP90x7A0vOgN6Y80PUAuC/FAXbXQF/zcoAhDj76AoPIg/p89UryshjEOfBx/ZatXkMUFLpcY9OYQNLlzIGPxIDMbZv7h7l9zOvnFAPHGrV5DtiyvK8hPtZ9+3Zf8fFyu8Rt4i8LulKtLwAAcBHxjCKwwqbncPPpncePHx89ryZ+W2SfuL5fLhLfRunn12pvnHz16lWZnsTbK73s6tWrZapn2KYD/vL59evXZeqyLlu9hujZtk8//bT5Nimfdfz60pjn4VzO2ptZXWctc3t1PTuo5zGnQVLzbTK5devWZBosNt/euXv3bu+zjzpmQ28mVbmmgV3zbZjLfuPGjTKt8T71MiAdV50H6jt37twp82VMn3HfAwDgoiNQBFaUX8ChwXwMLjTY12DdHCxtb2+31ovp2bNnZR0P9GPAIB5g1wKAHGCIXz5Te0GK81IgkjmwUUDgbXNg6DIsW73G+Pa3vz3Z3Nwsn7/zne+UN54el1+489FHH5WpKch0sKRg1wGQ929qG9VJPv744zJV/TRP6+Y2FQenmYI06QsUu4JJvyxGyzMdJwWWfUG789O6rqP/vIm5DXZ2dt6rl1PeBgCAC2v6gxHAivGteF0p3l7oWzb7km4z9K16Mel2vmmw1Jo3S97SdQujk+qS89E83yao/YvL4f2fdb2O44MPPjjK6+XLl83c2blta3VUcttZbR2neItmV35OcV3f8llLef/S1x+Ulwz1b+WR5WNWW6dv307KJ4rb1PIEAOC84ooisIJ0RWc6CC9XRrLpQLd1VURXYabnevPtfVq/dnVPdOXHV6xqlHetDNMB9dE+fZWwj28rjXy1z1eBXA5fSTzreh1HvCI5DRonP/jBD5pvs3Hb6kpnLq/K6iuqprLrylykfqT58RZN9S9dqZ0GjM2cQ9o2r6u2nIWOifLOdLxcn64rln10zFw35V879pqn/XRR3ZRPpLq63l39CQCA82ht+sPv+KMeAMDCKUDU7aUKfnKAg+PT31p88OBBCRIBALhoCBQBYMXoj93rpTKZrgDyjN3J0LOS6+vrBIkAgAuLW08BYMV03ZoZ3/KK4xm6tRkAgPOOK4oAAAAAgBauKAIAAAAAWggUAQAAAAAtBIoAAAAAgBYCRQAAAABAC4EiAAAAAKCFQBEAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWggUAQAAAAAtBIoAAAAAgBYCRQAAAABAC4EiAAAAAKCFQBEAAAAA0EKgCAAAAABoIVAEAAAAALQQKOJCuHfv3mRtba2VHj582CxdLO1X+79+/Xozp9/t27ffK7vm4VCtfZ48edIsXSz3s6Hjo2M/9vhjtajvxb6oPnFczrOrX6svxX0CAHASCBRxYV2+fLn5tFhPnz4t05s3b5ZpH5Xx8ePHzTeMdeXKlebTYo05ti9evJg8f/58cvXq1WYOloUDrnl/iaSgcHNzs/l26MGDB8f+pdSjR4/KtKtfv3nzpvk0mWxsbDSfAAA4HgJFXAj379+fHBwclOSB1K1bt8p00TToUzlUpj66evD27dvJ9vb2UdmdPHDEu/ZUunTpUknXrl1rli7Ws2fPSjnu3LnTzHnf69evy/SsflGB9/lKnAJ46Tt+XXS+KigU98ednZ3y3b9AmNerV6/KtKtfK1Dc3d0tn/kFBADgpBAoYuV5kOc0dNufBoMKJvr49tCYp1LtNrLarY/ZLLe+ermvTGjw6Xldt54tMwVELr9Sntd1C6auvMXtlPrqr/UVWPcFYLXjUNt/Xkd5x/mxj9WOf62cmqdlteOq5HLEeXG+5PLPG2z6ypmTy+Z6Rvn4qd9q+9x/xxyv3FY+n+L8bMz5NY94jov+T3CANw8Hl3F7z3OgNysfJwewLq9TbF8HozpesZ909ZHa8fLxkNzXJH6P6wIAzqnpDzVgJW1tbWlEVk0bGxvNWm37+/tlubbtom1zfk47OzvNWgcHu7u71XWctC/b3t7uXR7l9WLSPlfNdAB+VP74OSbNj/qOgdqyxseja7mPfS3FbWrr6bjneT4WtX5YU8sjJuVT23fsy3lfud2G9LWBUjRU3tgXxx6vvF5tHz43Zzm/ZpHbsKu/zMLt6rJb1/yxus4Xp9gGrldtm7z/vuPldZ2fUy3feY8BAGA1EChiJWlwp4FKHESbBzS1QYwHpprW1IIND/ZinrV55rJ1DUC1bMwAX+uMWW8VxEF/bhfP9zHxADUfWx+7rjZxu8cAZkjteFs8xkoeQHs/mdev9clozHH1PnK5uuaP5baNfdZ5xnLH45X7dy7/PMdL6zp/JXH7abvY9nn/87aB83OapZ8McV3dh61r/izG5qF21nqxvX0c3XfFxyvOM81XstgP4v6dx0m2IQBg+RAoYiV58NKXaoOYoQFOHKDGlAelzqcv1QZizr+2LPJ6Q0HHqvDgvjbYjQPh2sA2coBR42U5sIhygOJUK1ccJI8ZEHv9vgBm7PF3XvH41+bNyschp9xmfe0cuUyzHi8HNV3bab6W96WhNqzJ9e/rK7NweXN+XfNnMTYPrZP7hs8t90nloe9DyfvqOm/H9g8AwGrjGUWsnNrzXzW1NwT6WaGuF9noZRHTQVLz7R09TxafFRvzcoras0F7e3tlOvTG07HrrQq/lbHWJj4mWuZ2jc8ARnpWazpIbb61aR/TAKT6wg8/g+dnvbIbN240n95xWaaD5FEvPvL6fcds7ItstD/VReX1c4N+3u2zzz4r03noBUrTwX/z7Z319fXWM4d97RzNc7xUHz1LqvldL2Wa9/wa4pdamep9Es/adb1sRm8s7uqTYw29yEb8f2Lue7lP+v+VId5X13k7tn8AAFYbgSJWjgNADcA06OtKtYGVBjjarsYvd9DgUcFizEvbxFfQe+C0u7vbWi+m2ltNPXCrBSaRB4fnJVB0vfPbJBVgKAjXoFPBkd/YmP/EgI+NfPzxx2UaOfjoCh6031p/0TylWl/x8R46VuY69gWVeeAurlv8RYTEwb3aSfVT0Dpv0OGXk6gMsQ3Uh8V9zmKQGsWyznO8HCz39e15z6+xtL0DZr9U6DjcDjHYdp75pT+zqgVlfsGPA0T31fzG09wn3a7KL7dnTFbr011BKQDgHJr+UABWznSgU2596kq+Lc23bfUl31aVb0vLybdviW/p6ksydKtXzFNq6zj5drBV4zboOhbTQK1Z81BtHad4C9xQH1DSrZFSW5aT1h06XpHWr63jFMtqff0x94Wcv+p7HGrnmF9OsX8NnTexrLXlTmOPV26rsefXceXjnY/BWF3lnTc/G9Mf83FVO9f6ptt4qB+ozF31yf0in7sAgPOFK4pYSfp7ddNBS/OtbTrQmevvDOq38spT22fTc6V1BUNXqKaDseZb23TwVNYXXz0Zo3b15jyIV5Bym00HpK0rtaK2UxtG08FvmZ+vSI6V96vvteM3y/HK5R5D/bKrbvkKWb4yqT5/HLoiqTrn/avPa//xSqXK2XV+5bKOPV6ztNfY8+u4VGfl5brO+zcIVd78/4bKf5yrnqLy6RzJlHdXG6gOtbb21UQt0/Gp0b76yqxbaQEAF8fa9IfNyfzEBYAK3fKo2/s0uO27LRPv0y2GGpxrAD9vkAwAADAPAkUAp0LPMuVn10RXM457dey8cmBYoytWx71CBQAAMBa3ngI4FV23Gs57e99FkF8oE9FuAABgkbiiCAArRM+a6XnDsbjlFwAAzIMrigCwQmYJEqX290QBAACGcEURAAAAANDCFUUAAAAAQAuBIgAAAACghUARAAAAANBCoAgAAAAAaCFQBAAAAAC0ECgCAAAAAFoIFAEAAAAALQSKAAAAAIAWAkUAAAAAQAuBIgAAAACghUARAAAAANBCoAgAAAAAaCFQBAAAAAC0ECgCOLYnT55M1tbWynSMe/fulfVv377dzFlOs9YL4z18+LC07YsXL5o5/dRXtL76DgAAOH0EigCO7dGjR2V65cqVMh3y9OnTMr1582aZLqtZ64XxPv/88zK9du1amQ5ZlT4DAMB5sXYw1XwGgLlcv3598vz588l5++9kEfX66quvJt/4xjeab5PJ3t5e2e95d/ny5TJ98+ZNmQIAgOXCFUVgxfn2SCcFGbo9r3Zbp2/3iyneVqltPV+fc975Fkyvr2BK4rp5fd862LU878vrK6CI29ZuPTzLeh3H1772taMg8eXLlyUg1b7HqtUj17Umb6ekNqxR+3sdHYfY1g72xLcTx1Tbv/N7+/ZtSXmbeCtqrEttucTy5XW65kutvH23wbosXW0KAMC5oyuKAFbT1taWLnVV0/b2drPWodo6Tjs7O2WdS5cuHc2Ln+O8qLZOTPv7+82a9bJGKkNe3pWi2nKnRdRrHl9++eVRXpubm83c2eU2q5XbbWB9dYt9ZszxcLupPWrLlWbph0rRxsZGa1k+ThKXO/n41Ob3lVVpd3e3bJt5eW5PAADOKwJFYEV5wKsALKoNeB0c5EGu1tF8DchNA3vn4QG3OI/MAcXYAbTLHfdpcRAvDi69rsoQg4VlqtdYCgxdjr29vWbu8Ti/2Bdc/hioOfDKwZvb3W3r9lOKbRXbcIjzyPsS5xP7aB/nlfu6eXnsG6J5SpHn5X27vbr2AQDARTP80x7A0vKgN6Y80PUAuC/FAbaDiRggiObVgjsHc3n9Ln0BRF4WgzgHM67fstVriIJCl2tsGhNIdgVRORjzekNJutpKND8HZOJtcqoF2l53LB/rWl7mvuIy+/jFbdwmfanWFwAAuIh4RhFYYdNzuPn0zuPHj1vPjvltkX3i+n65SHwbpZ/Jq71x8tWrV2V6Em+v9LKrV6+WqZ5hmw74y+fXr1+Xqcu6bPUaomfbPv300+bbpHzW8etLY56Hc71iXcXtc+vWrTId017TYKtMlac+57r7OcbYXn7e0c9zZjdu3Gg+veP8x3LZa3mZy6SXAalMOg/Ud+7cuVPmy5g2cN8DAOCiI1AEVpRfwLG7u9sKLjY2NkqAZQ4gtre3W+vF9OzZs7KOXuahbZVH5AF2LaBSgJDX98tnHIhFOYCJHPQoIPC2OTB0GZatXmN8+9vfnmxubpbP3/nOd8obT4+rqw65/A6AFDzV2krJ7S9qr/iCG32+e/du+Rz3pUBMQV/OS/NqwaaPRS6vXxZTe6GM6ljLK3J+Wtdt7D9vYm6DnZ2d98rrlLcBAODCmv5gBLBifCteV4q3IfqWzb6k2xJ9q15M08H5e7cUzpK3DN3yqLrkfDTPtwlq/+JyeP9nXa/j+OCDD47yevnyZTN3Nl1lrNVX60qen5Nv0+y6RdNt5vwkr1NLOg7W1x90XGSof8f8LLdHbZ2+fTvFukncppYnAADnFVcUgRWk2yKng/ZyZSSbDnRbV0V0FWZ6rjff3qf1a1f3RFd+um4pFOVdK8N0QH20z3iVqotvK418pcxXgVwOX0k863odR7wiOQ0aJz/4wQ+ab8enWy67qOy6Mlejuvk2zfv3779Xf23rYxmv7Gm7SN/zvEjHRFeBMx0v5+/bfmehMrluyr927DVP++miOsa6Sbxi2tWfAAA4j/iD+wCAUXRr6NbW1oW5PVN/a/HBgwcn8ssBAABWDYEiAKBFzwmur68339ouyo8MtwE/IgEAFxW3ngIAWmq3AutW54sUNA3d2gwAwHnHFUUAAAAAQAtXFAEAAAAALQSKAAAAAIAWAkUAAAAAQAuBIgAAAACghUARAAAAANBCoAgAAAAAaCFQBAAAAAC0ECgCAAAAAFoIFAEAAAAALQSKAAAAAIAWAkUAAAAAQAuBIgAAAACghUARAAAAANBCoAgAAAAAaCFQBAAAAAC0ECgCAAAAAFoIFHEh3Lt3b7K2ttZKDx8+bJYCAAAAiAgUcWFdvny5+QQAAAAgWjuYaj4DF8L169cnz58/n9D1AQAAgDquKGLl5VtKb9++3SypU5B46dKl5ludbkvNeSrpFtZM8+O6SgAAAMAqI1DEynKAlj1+/LhcNax58eJFmd68ebNMa7Tt3bt3m2+HlKfS1atXmzmTyZMnT8r+NT/TfO8LAAAAWDUEilhJurKnAG1jY6PcQhqTrhbqqmEtUNvb2yvTrkBRwZ+23d7ePspvf3+/WTqZ3Lhxo0yV9+bmZvms5XH/2la++OKLMgUAAABWDYEiVtKDBw/KVEGdb/d0evv2bVn2+vXrMo2ePn1apl0vsrly5UqZKn/nt76+fhQ4Xrt2rSz/3ve+V6ai5XH/LtubN2/KFAAAAFg1BIpYObrqN4aDvujVq1dleuvWrTLNFAjGK4im4C/ezuqAsw9vVQUAAMCqIlDEynEAqFtM4y2fOfnqX9T3IhvdTuoriPl2Um0TrxA6CNzd3W2tF9P9+/fLOgAAAMCqIVDEylEAqGcTdYtpvOUzJr/5NL+RVPJ2/sP78ZnCfDuptnGe8q1vfatM9ZxiXC8mAAAAYFURKGIlPXv2bLK1tdV8a9PzhI8ePWq+jacrhsrTL6OJ8hXCO3fulKuJNb7SCQAAAKwq/uA+AAAAAKCFK4oAAAAAgBYCRQAAAABAC4EiAAAAAKCFQBEAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWggUAQAAAAAtBIoAAAAAgBYCRQAAAABAC4EiAAAAAKCFQBEAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWggUAQAAAAAtBIoAAAAAgBYCRQAAAABAC4EigGN78uTJZG1trUzHuHfvXln/9u3bzZzlNGu9LroXL16U9nr48GEzZ3Hcp2I6i3IAAHBeECgCOLZHjx6V6ZUrV8p0yNOnT8v05s2bZbqsZq3XRffFF1+U6eXLl8v0rC1LOQAAWEUEigCO7dWrV2V67dq1Mh3y7NmzycHBweTOnTvNnOU0a73m8dVXX7WugqltVtWbN2/K9CwC6/v375c+pbSxsVHm3bp1q0wBAMDsCBSBFefbI52uX79ebsOr3dapW/Hiukrxtkpt6/n6nPPOt2B6/efPn5fvcd28vsrTtzzvy+vrqlDcVnXLzrJex/G1r31t8o1vfKN8fvnyZQlytO+xavXIda3J2ynF2zTjfF+Vi/kqxfV9fB4/fly+r6+vV9f1rakx5fn5+MZ1nYbaX8ft0qVLzbdusY8BAIBkOjABsKK2trYOdBrX0vb2drPWodo6Tjs7O2Wd6eD6aF78HOdFtXVi2t/fb9aslzVSGfLyrhTVljstol7z+PLLL4/y2tzcbObOLrdZrdxuA+urm/tMnKfjJhsbG635u7u7Zb7kZTl5XU3jfJVF+4zzlNy+ff27i7bVcpe7j9sinysAAGA6HmimAFZM14DYA+k4kPeAOAcNHrhroG9x4B4DIueROVjJeXdxueM+zcu8HwcKXldlULJlqtdYCgxdjr29vWbu8Ti/2Bdc/hgEOaDLgZHbPbZtV/2dR+Y2HxN05YDR26j8sQw1Xfu30zpuAABcNN0/bQEsvTjYdsqBowfOfSkOzj0Qj8GUaF4MvMzBXF6/S19AkZfFIM7BjOu3bPUaoqDQ5RqbxgSSbrN83NWGmq/lkoOzrmRd9de8WjCX99cnHru+9u0qc18w6XKPKQcAAOjGM4rACpuew82nd/ScWHzmym8Y7RPX9wtJ4gtc/ExY7S2ls77wpe+Np1529erVMn379u1kOvAvn1+/fl2mLuuy1WuInvH79NNPm2+T8lnHry+NeV7R9crP2bl9/EKXMe01DcCaT+/Wj/XXs4RSa69ZXmTjvKfBXGf7qu6bm5vNt7a+t+X6uPEiGwAAjodAEVhRfrGHBtsxuNjY2CgBljmA2N7ebq0Xk9+0qUBA2yqPyAP72gBdLw7J6/slIbWXjuQAJnKwcePGjaNtc2DoMixbvcb49re/fRT8fOc73ylvPD2urjrk8jv4VuBdayslt7/U2mtvb69Ma+2lcijQjIGf/7ZhfPGNDAVzWl/l39nZaZVP36W2fxv7IhsAADBg+sMXwIoZuu0y3oboWzb7km7T8y17MekWv2mw0Jo3S97Sdfugk+qS89E838qo/YvL4f2fdb2O44MPPjjK6+XLl83c2XSVsVZfrSt5fk5qdxnTtj4uVlvHSfn19QMdj8jHvi95/7X65uR6WaxfXgYAAA5xRRFYQboiMx0oH11hiaaD4KM/FC+6wjM915tv79P6XVd2dFVJV2i6KO9aGXyVU+JVqi6+rTTylTJfCXM5fCXxrOt1HPGK5DRonPzgBz9ovh2f/0RFjcredbVNdfPftVT9daU2UnvmeZG2z9SG2qfyG9MPTH8TUf3b9Fn5xHnH8dFHH5Wp8lz2v+UJAMBZWZv+oDz+qAcAgBWg21rv3r1bAl8FsAAAoI5AEQBwYeiZSYJEAACGESgCAAAAAFp4RhEAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWggUAQAAAAAtBIoAAAAAgBYCRQAAAABAC4EiAAAAAKCFQBEAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWggUAQAAAAAtBIoAAAAAgBYCRQAAAABAC4EiAAAAAKCFQBEAAAAA0EKgCAx48eLFZG1t7Sjdvn27WfJOXGeMe/futfJUun79erMUAAAAOFsEisCMLl++3Hx6Z29vr0w3NjbKtI8CwgcPHjTfAAAAgOVDoAgMuHbt2uTg4GCyvb1dvt+8ebNMozt37pR1nj171syp05XH58+fT7a2tsr6MQ1tCwAAACwKgSJWXr6FM94a+vDhw9YyBWq6IlhbN9L8uJ28efOmTG/dulWmT548aa2j1JWf97m+vl6+P378+GgblREAAABYJgSKWFkO5jIFYX7e79WrV2VqH3744eTt27fNt8N1FfCZnzXU/EiBnvK6dOlSM6eudrURAAAAWDUEilhJehmMgjk9E5hv4VQwp9s7FfTdv39/sr+/32x1GFx6Pd9K6iuF4it+Ozs7R+tpewWXyjMGgrqy6HX8bOKNGzfKNNM+4nreTkm3rQIAAADLhEARK8kvg1Hw5ls4nXzF8PXr12XqF80oMFTgaA4QHdwp+JTd3d1W8KZnFPueTxTlpQBV6/ZReYeuSgIAAABnjUARKyfeKtrnypUrZerbT69evVqm9vTp01Zwp+/iZxAjB5W1N57qyqWC06HbTl1ubk8FAADAsiNQxMpxAKggL97CmVMOAONtoX3BXX65jG5X9TOLtSDSVyyHAkAHmwSKAAAAWHZr0wH1QfMZWBl6WY1u4+yiPz+hAG9zc7OZ846W5ZfV6DRQgHj37t1mzjt6rjDuS98V9MWX4kRaHv/Uha5Cdq2r21xrwScAAABwlriiiJWkQEwBX42eJ3z06FHz7X05SDQ9l6jALVJeH3/8cfPtkG5h7Qr8amZZFwAAAFgGXFEEAAAAALRwRREAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWggUAQAAAAAtBIoAAAAAgBYCRQAAAABAC4EiAAAAAKCFQBEAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWggUAQAAAAAtBIoAAAAAgBYCRQAAAABAC4EiAAAAAKCFQBEAAAAA0EKgCAAAAABoIVAEAAAAALQQKGLQ2traUbp+/Xoz931xPacnT540SwEAAACsCgJFzOTq1avNp3cUDCooBAAAAHA+EChi0MHBwWRnZ6d8vnnzZplGjx49KtP9/f2ybky3bt0qywAAAACsDgLFFXf58uXWrZ7x1tAXL160lj18+HBy+/bto+/atubevXut7ZTPq1evyrIbN26UqfLy8sePH5d56+vr5fvYfPN6ubxKeb7yAAAAAHC6CBRXlIOut2/fNnMOPX/+/CgAe/36dZnaJ598chTUibbNgZfyfPDgQfPtkALAN2/elM/Xrl0r01k40Mv5av+ar+WSy3vp0qVSPu3flIfXBwAAAHA6CBRXkK7mKWBSIJVv9dzY2CgBmJ4b1G2fmqf1RAGk19vd3S3zHACKA8zt7e2j9ZREAabytjt37rTyidvEPMWBnm5f9TpKzu+LL74oU5fXeaoeqqfz3traKnWZJ1gFAAAAMB6B4grSlUHxFbmYdEVRHKzp6pvWU1D27NmzMk+83M8cKvjUegrm7t+/X+ZZ3/OJT58+LdPaS25Et7qKgj8Fl9Fnn31WpjmwjN/13KPLo2ch87oAAAAATh6B4opx4Dck336aAzkHeH7mMH+P/HxiLRjs2060XFcBay+12dvbK9McgDpPBZdcPQQAAAAWj0BxxcTAKd7GmZMDMwddtWCsdhunbwM1PSPoZwtrwaCu8PXdDqqAVYGtryya8r17927ZNl9pdGDKG1MBAACAs7E2DSoOH0LDylDQFV9Kk/k2U92Kmuk5v7ytnwnc3Nws00h5+XZWUWD38ccflyCvJt/iqltau9YVdz89U1nbv2if3HIKAAAALA5XFFeQntXTC15qFAjGQC3rCjB19U7PA0bKy88Rmq4Q+orfGLpa6EA08ottjEAQAAAAWB5cUQQAAAAAtHBFEQAAAADQQqAIAAAAAGghUAQAAAAAtBAoAgAAAABaCBQBAAAAAC0EigAAAACAFgJFAAAAAEALgSIAAAAAoIVAEQAAAADQQqAIAAAAAGghUAQAAAAAtBAoAgAAAABaCBQBAAAAAC0EigAAAACAFgJFAAAAAEALgSIAAAAAoIVAEQAAAADQQqAIAAAAAGg5N4Hiw4cPJ2tra0dJ32vu3bvXWk/p+vXrzVIAAAAAwLm9onj58uXm0zsKCB88eNB8AwAAAADUrB1MNZ/PBQWDz58/n+RqvXjxYrK+vj7Z2tqaPHr0qJkLAAAAAMgWckXxyZMn793uGW8NvX379tF8XQlUUNe1bqR1vY5vH33z5s1kY2OjfBavoyBRHj9+PFO+SrpdNcq3r6r8eb7qAAAAAACr6NQDRQVdm5ubzbd37t69exSAvXr1qkzNQZ1p3cjPI759+7aZMylXERWwad7Vq1ebubNxoBfzFd2uGp9jVDAaqY5K8bbWjz76qPkEAAAAAKvlVANFBVcKura3t8utoE77+/tluW8Bffbs2WRnZ6d81vq7u7tH6/rqoK/Q6eqkA0fl4/W0va4Wys2bN8tUFNTFfLy+0p07d8o8UfDpQC+u47IqEHUZVG4t022sou1UbpdHYhkAAAAAYJWc2jOKCuhqVxIz715XAxXoKUi8detWmSe6Uie+itf1DKJ4mQK2a9euNXMP6UrhpUuX3rsaaH3LdaVRwWAum/enIFTBLgAAAACcB6d2RfHp06fNp24KzMy3n165cqVMRVfwdKUuXp1zYFbjIC8HiQpapesqn5f7WcPMdYlBoqgsqgNBIgAAAIDz5NQCRT8nqNszdfWvluLVOwddMcjb29sr0xzgad38shhdeVRQWQsivZ+uQNHBqa4aOmg05av96fbZyPvnFlMAAAAA582p/nkM3c7ZR88V6kpifAmMKcD0M4fi2zt9i2qm5QroTIGdniVU8FjTdRtpTfyTGl37F+3z/v37zTcAAAAAWE2n+jIbxaDx9tJIgVp8mUzWFYwpYMtX95RXvrKnK5pdQWKNglC/nCZSHRwkSn5DKwAAAACcN6d6RREAAAAAsHpO/e8oAgAAAABWC4EiAAAAAKCFQBEAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWggUAQAAAAAtBIoAAAAAgBYCRQAAAABAC4EiAAAAAKCFQBEAAAAA0EKgCAAAAABoIVAEAAAAALQQKAIAAAAAWtYOpprPvb773e82nwAAAAAAi7bImGx0oAgAAAAAuBi49RQAAAAA0EKgCAAAAAAIJpP/HwX8m36xJVh8AAAAAElFTkSuQmCC\"\u003e\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003eFinally, we merged the data frames with the COVID-19 time series to track sentiment changes over two periods of time and to examine any correlations between the epidemiological situation in Morocco and the variation in sentiment results. Figure \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e shows the overall workflow of our proposed approach.\u003c/p\u003e\n\u003c/div\u003e\n\u003cp\u003e[\u0026Dagger;] C0VID-19 time series data repository (accessed 25/5/2024): https://github.com/CSSEGISandData/COVID-19\u003c/p\u003e\n\u003cp\u003e[\u0026sect;] \u003cstrong\u003eHespress\u003c/strong\u003e dataset repository (accessed 26/5/2024): https://github.com/HankarM88/Hespress_COVID-19_Dataset\u003c/p\u003e\n\u003cp\u003e[**] TBCOV dataset (accessed 26/5/2024): https://crisisnlp.qcri.org/tbcov\u003c/p\u003e\n"},{"header":"4. Results and discussion","content":"\u003cp\u003eThis section outlines the experiment results of applying our proposed approach. While most approaches in the literature review have examined the question of COVID-19 pandemic effects on people by simply applying baseline sentiment analysis methods to identify sentiments in the user-generated content from Twitter, Facebook, and other social media outlets; we combined many techniques including topic modeling, word embeddings, sentiment analysis, and time series to realize this work. Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e below shows the frequency of sentiment polarities and the polarity rate regarding each topic.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003cdiv class=\"colspec\" align=\"char\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003ctable id=\"Tab4\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003etopic-level sentiment analysis results\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eTopic\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003ePolarity\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eFrequency\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003ePercentage (%)\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"3\" align=\"left\"\u003e\n\u003cp\u003eEconomy\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePositive\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e254\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6.3\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNeutral\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e537\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e13.3\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNegative\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e3285\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e80.5\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"3\" align=\"left\"\u003e\n\u003cp\u003eEducation\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePositive\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e480\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8.8\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNeutral\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e849\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e15.6\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNegative\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e4114\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e75.6\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"3\" align=\"left\"\u003e\n\u003cp\u003eFear\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePositive\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1550\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e17.09\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNeutral\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1118\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e12.4\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNegative\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e6400\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e70.6\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"3\" align=\"left\"\u003e\n\u003cp\u003eHealth\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePositive\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e631\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8.85\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNeutral\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1198\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e16.8\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNegative\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e5300\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e74.5\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"3\" align=\"left\"\u003e\n\u003cp\u003eSupport\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePositive\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e344\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e7.7\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNeutral\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e524\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e11.7\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNegative\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e3604\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e80.6\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"3\" align=\"left\"\u003e\n\u003cp\u003eVaccine\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePositive\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e283\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3.8\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNeutral\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1252\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e17\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNegative\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e5815\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e80.2\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"BlockQuote\"\u003e\n\u003cp\u003eThe overall findings showed that COVID-19 outbreak has dramatically affected most aspects of normal life in Morocco, meaning that people had reacted with high rates of negative sentiments regarding all COVID-19 related topics. For example, more than 80% of respondents were feeling very deficient on how the outbreak has calamitously affected the economy and job market, 70% expressed feelings of anger, fear, and anxiety towards the novel coronavirus spread, and 80% of them expressed negative feelings about education, especially during lockdowns where students had to stay at home and learn online. People have worried about health systems collapse because of the huge number of new cases received every day at hospitals. The majority of Moroccans reacted to vaccine campaigns (80%) and expressed concerns of worry and hesitancy about the ones that the health authorities suggested for people to take, especially Sinopharm and AstraZeneca vaccines.\u003c/p\u003e\n\u003cp\u003eIn the first year of the pandemic in Morocco (2020), people have been mostly concerned about health and education. This led them to express emotions of fear and disruption as illustrated in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e. However, high rates of negative concerns and reactions have emerged about the vaccine in the second year of the pandemic (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e). The widespread dissemination of misleading information and fake news about COVID-19 outbreak and the vaccines through social media and the internet has led to a growing sense of fear and distrust among people, making them hesitant to take the vaccine. This can be seen clearly in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e which shows the impacts of COVID-19 regarding all other topics except vaccine. sentiments of fear were highly negative for these topics. Consequently, some people have formed groups of anti-vaccination to manifest against vaccines, and imposing mandatory vaccine passes for accessing government buildings, and other public places. As a result, vaccine hesitancy has become a pressing issue that drives people to delay in accepting or refusing vaccines despite their availability. It is often fueled by concerns, doubts, and misinformation shared on social media and the internet. This hesitancy can lead to lower vaccination rates, hindering efforts to achieve herd immunity and control the spread of the virus.\u003c/p\u003e\n\u003c/div\u003e\n\u003cp\u003eWe intended to separate our study into three different periods of the pandemic to track sentiment changes and compare these variations from March 2020 until the end of 2022 in the history of the pandemic. Figure\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e shows that negative sentiment rates started to increase exponentially from March 2020, the very beginning of the outbreak, to peak in November 2020. After this period, people\u0026rsquo;s general concerns about the COVID-19 situation started to fade. In the second period, negative sentiment feedback was not as high as in 2020. However, it is obvious from Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e that negative sentiments in the time window of March 2021 to May 2021 have spiked and after that started to decrease. After June 2021, negative polarities started to increase slightly until November and decreased at the end of the year. In the beginning of 2022, negative sentiments continued to decrease until March and then started to increase and peaked in June and started to decrease until the end of the year while rates of positive sentiments slightly increased in the last months of the year.\u003c/p\u003e\n\u003cp\u003eThe variation of negative sentiments over time is well explained in Figs.\u0026nbsp;10, \u003cspan class=\"InternalRef\"\u003e11\u003c/span\u003e, and 12. The epidemiological situation, represented by monthly records of COVID-19 cases and fatalities, directly influences how people react to the pandemic. The line plots show that changes in COVID-19 situations directly affect feedback sentiments during a given period of time. High rates of negative sentiments are explained by the spikes of coronavirus spread (cases and fatalities) among people and the subsequent actions taken, such as lockdowns, wearing masks, social distancing, and other protective measures to slow down the outbreak.\u003c/p\u003e\n\u003cp\u003eAs the infection rate decreases and the government slightly eases protective measures, negative sentiments also decrease, and people gradually return to their normal lives. When the situation improves, and there is a lower risk of COVID-19 transmission, the negative sentiments may start to subside. People might feel more relieved and less anxious as they perceive a lower threat level. As a result, they may gradually return to their pre-pandemic routines and lifestyles. The interplay between the epidemiological situation, government actions, and people's reactions can significantly impact the changes in sentiments overtime during the course of the COVID-19 pandemic.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eIn this paper, we proposed an approach that combines many different NLP techniques to examine COVID-19's effects on Moroccan citizens through their generated content in social media and other web platforms. The system takes textual data from different sources (Twitter and the online newspaper Hespress) and performs necessary text cleaning and preprocessing, then extracts main topics within the corpus using the BERTopic model, and finally identifies sentiments on a topic-based level using the pre-trained Arabic CAMeL analyzer. The results showed that the pandemic has drastically affected all aspects of normal life. We also used COVID-19 times series data of monthly cases and fatalities to examine sentiment variation within three different periods in the outbreak\u0026rsquo;s history, and we found that the epidemiological situation highly impacts sentiment analysis results.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eHaleem A, Javaid M, Vaishya R (2020) Effects of COVID-19 pandemic in daily life. Curr Med Res Pract 10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cmrp.2020.03.011\u003c/span\u003e\u003cspan address=\"10.1016/j.cmrp.2020.03.011\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOnyeaka H, Anumudu CK, Al-Sharify ZT, Egele-Godswill E, Mbaegbu P (2021) COVID-19 pandemic: A review of the global lockdown and its far-reaching effects. Sci Prog 104. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/00368504211019854\u003c/span\u003e\u003cspan address=\"10.1177/00368504211019854\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eŠkare M, Soriano DR, Porada-Rochoń M (2021) Impact of COVID-19 on the travel and tourism industry. Technol Forecast Soc Change 163. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.techfore.2020.120469\u003c/span\u003e\u003cspan address=\"10.1016/j.techfore.2020.120469\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePriya SS, Cuce E, Sudhakar K (2021) A perspective of COVID 19 impact on global economy, energy and environment. Int J Sustain Eng 14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/19397038.2021.1964634\u003c/span\u003e\u003cspan address=\"10.1080/19397038.2021.1964634\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChang AY, Cullen MR, Harrington RA, Barry M (2021) The impact of novel coronavirus COVID-19 on noncommunicable disease patients and health systems: a review. J Intern Med 289. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/joim.13184\u003c/span\u003e\u003cspan address=\"10.1111/joim.13184\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePokhrel S, Chhetri R (2021) A Literature Review on Impact of COVID-19 Pandemic on Teaching and Learning, Higher Education for the Future. 8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/2347631120983481\u003c/span\u003e\u003cspan address=\"10.1177/2347631120983481\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePassavanti M, Argentieri A, Barbieri DM, Lou B, Wijayaratna K, Foroutan Mirhosseini AS, Wang F, Naseri S, Qamhia I, Tanger\u0026aring;s M, Pelliciari M, Ho CH (2021) The psychological impact of COVID-19 and restrictive measures in the world. J Affect Disord. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jad.2021.01.020\u003c/span\u003e\u003cspan address=\"10.1016/j.jad.2021.01.020\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSerafini G, Parmigiani B, Amerio A, Aguglia A, Sher L, Amore M (2020) The psychological impact of COVID-19 on the mental health in the general population. QJM 113. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/qjmed/hcaa201\u003c/span\u003e\u003cspan address=\"10.1093/qjmed/hcaa201\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrooks SK, Webster RK, Smith LE, Woodland L, Wessely S, Greenberg N, Rubin GJ (2020) The psychological impact of quarantine and how to reduce it: rapid review of the evidence, The Lancet. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S0140-6736(20)30460-8\u003c/span\u003e\u003cspan address=\"10.1016/S0140-6736(20)30460-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNemes L, Kiss A (2021) Social media sentiment analysis based on COVID-19. J Inform Telecommunication 5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/24751839.2020.1790793\u003c/span\u003e\u003cspan address=\"10.1080/24751839.2020.1790793\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eQuyyam T, Ghous H (2021) Sentiment Analysis of Amazon Customer Product Reviews: A Review. Int J Sci Res Eng Dev. 4\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChakraborty K, Bhattacharyya S, Bag R (2020) A Survey of Sentiment Analysis from Social Media Data. IEEE Trans Comput Soc Syst 7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/TCSS.2019.2956957\u003c/span\u003e\u003cspan address=\"10.1109/TCSS.2019.2956957\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKasri M, El-Ansari A, El Fissaoui M, Cherkaoui B, Birjali M, Beni-Hssane A (2023) Public sentiment toward renewable energy in Morocco: opinion mining using a rule-based approach. Soc Netw Anal Min 13:124. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s13278-023-01119-3\u003c/span\u003e\u003cspan address=\"10.1007/s13278-023-01119-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBirjali M, Kasri M, Beni-Hssane A (2021) A comprehensive survey on sentiment analysis: Approaches, challenges and trends. Knowl Based Syst 226:107134. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.knosys.2021.107134\u003c/span\u003e\u003cspan address=\"10.1016/j.knosys.2021.107134\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBirjali M, Kasri M, Beni-Hssane A (2021) A comprehensive survey on sentiment analysis: Approaches, challenges and trends. Knowl Based Syst 226. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.knosys.2021.107134\u003c/span\u003e\u003cspan address=\"10.1016/j.knosys.2021.107134\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChandrasekaran R, Mehta V, Valkunde T, Moustakas E (2020) Topics, Trends, and Sentiments of Tweets about the COVID-19 Pandemic: Temporal Infoveillance Study. J Med Internet Res 22. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2196/22624\u003c/span\u003e\u003cspan address=\"10.2196/22624\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXue J, Chen J, Chen C, Zheng C, Li S, Zhu T (2020) Public discourse and sentiment during the COVID 19 pandemic: Using latent dirichlet allocation for topic modeling on twitter. PLoS ONE 15. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pone.0239441\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0239441\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBoon-Itt S, Skunkan Y (2020) Public perception of the COVID-19 pandemic on twitter: Sentiment analysis and topic modeling study. JMIR Public Health Surveill 6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2196/21978\u003c/span\u003e\u003cspan address=\"10.2196/21978\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYin H, Song X, Yang S, Li J (2022) Sentiment analysis and topic modeling for COVID-19 vaccine discussions. World Wide Web 25. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11280-022-01029-y\u003c/span\u003e\u003cspan address=\"10.1007/s11280-022-01029-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eQorib M, Oladunni T, Denis M, Ososanya E, Cotae P (2023) Covid-19 vaccine hesitancy: Text mining, sentiment analysis and machine learning on COVID-19 vaccination Twitter dataset. Expert Syst Appl 212. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.eswa.2022.118715\u003c/span\u003e\u003cspan address=\"10.1016/j.eswa.2022.118715\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAbdul-Mageed M, K\u0026uuml;bler S, Diab M (2012) SAMAR: a system for subjectivity and sentiment analysis of Arabic social media, in: 12 Proceedings of the 3rd Workshop in Computational Approaches to Subjectivity and Sentiment Analysis\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZarra T, Chiheb R, Moumen R, Faizi R, Afia AE (2017) Topic and sentiment model applied to the colloquial Arabic: A case study of Maghrebi Arabic, in: ACM International Conference Proceeding Series, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/3128128.3128155\u003c/span\u003e\u003cspan address=\"10.1145/3128128.3128155\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShelke N, Deshpande S, Thakare V (2017) Domain independent approach for aspect oriented sentiment analysis for product reviews. Adv Intell Syst Comput. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-981-10-3156-4_69\u003c/span\u003e\u003cspan address=\"10.1007/978-981-10-3156-4_69\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMadani Y, Erritali M, Bouikhalene B (2023) A new sentiment analysis method to detect and Analyse sentiments of Covid-19 moroccan tweets using a recommender approach. Multimed Tools Appl. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11042-023-14514-x\u003c/span\u003e\u003cspan address=\"10.1007/s11042-023-14514-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHankar M, Birjali M, El-Ansari A, Beni-Hssane A (2022) COVID-19 Impact Sentiment Analysis on a Topic-based Level. J ICT Stand 10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.13052/jicts2245-800X.1027\u003c/span\u003e\u003cspan address=\"10.13052/jicts2245-800X.1027\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHankar M, Birjali M, El-Ansari A, Beni-Hssane A (2022) Arabic Topic Modeling-Based Sentiment Analysis on COVID-19 Feedback Comments, in: Lecture Notes in Networks and Systems, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-3-030-91738-8_9\u003c/span\u003e\u003cspan address=\"10.1007/978-3-030-91738-8_9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHegazi MO, Al-Dossari Y, Al-Yahy A, Al-Sumari A, Hilal A (2021) Preprocessing Arabic text on social media, Heliyon. 7 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.heliyon.2021.e06191\u003c/span\u003e\u003cspan address=\"10.1016/j.heliyon.2021.e06191\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAbu Farha I, Magdy W (2021) A comparative study of effective approaches for Arabic sentiment analysis. Inf Process Manag 58. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ipm.2020.102438\u003c/span\u003e\u003cspan address=\"10.1016/j.ipm.2020.102438\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eImran M, Qazi U, Ofli F (2022) TBCOV: Two Billion Multilingual COVID-19 Tweets with Sentiment, Entity, Geo, and Gender Labels, Data (Basel). 7 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/data7010008\u003c/span\u003e\u003cspan address=\"10.3390/data7010008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBlei DM, Ng AY, Jordan MI (2003) Latent Dirichlet allocation. J Mach Learn Res 3. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/b978-0-12-411519-4.00006-9\u003c/span\u003e\u003cspan address=\"10.1016/b978-0-12-411519-4.00006-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHoyer PO (2004) Non-negative matrix factorization with sparseness constraints. J Mach Learn Res. 5\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRudkowsky E, Haselmayer M, Wastian M, Jenny M, Emrich Š, Sedlmair M (2018) More than Bags of Words: Sentiment Analysis with Word Embeddings. Commun Methods Meas 12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/19312458.2018.1455817\u003c/span\u003e\u003cspan address=\"10.1080/19312458.2018.1455817\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKasri M, Birjali M, Beni-Hssane A (2019) A comparison of features extraction methods for Arabic sentiment analysis, in: Proceedings of the 4th International Conference on Big Data and Internet of Things, ACM, New York, NY, USA, : pp. 1\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/3372938.3372998\u003c/span\u003e\u003cspan address=\"10.1145/3372938.3372998\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKasri M, Birjali M, El Ansari A, Beni-Hssane A (2022) Enhanced Word Embeddings with Sentiment Contextualized Vectors for Sentiment Analysis, in: : pp. 77\u0026ndash;86. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-3-030-91738-8_8\u003c/span\u003e\u003cspan address=\"10.1007/978-3-030-91738-8_8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKasri M, Birjali M, Nabil M, Beni-Hssane A, El-Ansari A, Fissaoui ME (2022) Refining Word Embeddings with Sentiment Information for Sentiment Analysis. J ICT Stand. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.13052/jicts2245-800X.1031\u003c/span\u003e\u003cspan address=\"10.13052/jicts2245-800X.1031\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDevlin J, Chang MW, Lee K, Toutanova K (2019) BERT: Pre-training of deep bidirectional transformers for language understanding, in: NAACL HLT 2019\u0026ndash;2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies - Proceedings of the Conference\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLe Q, Mikolov T (2014) Distributed representations of sentences and documents, in: 31st International Conference on Machine Learning, ICML 2014\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEgger R, Yu J, Topic Modeling Comparison Between A (2022) Top2Vec, and BERTopic to Demystify Twitter Posts. Front Sociol 7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fsoc.2022.886498\u003c/span\u003e\u003cspan address=\"10.3389/fsoc.2022.886498\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAbuzayed A, Al-Khalifa H (2021) BERT for Arabic Topic Modeling: An Experimental Study on BERTopic Technique. Procedia Comput Sci 189:191\u0026ndash;194. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/J.PROCS.2021.05.096\u003c/span\u003e\u003cspan address=\"10.1016/J.PROCS.2021.05.096\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAntoun W, Baly F, Hajj H (2021) AraBERT: Transformer-based Model for Arabic Language Understanding\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eObeid O, Zalmout N, Khalifa S, Taji D, Oudah M, Alhafni B, Inoue G, Eryani F, Erdmann A, Habash N (2020) CAMeL tools: An open-source python toolkit for arabic natural language processing, in: LREC 2020\u0026ndash;12th International Conference on Language Resources and Evaluation, Conference Proceedings\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Topic modeling, Time series, Sentiment analysis, COVID-19 impacts, Twitter, Hespress","lastPublishedDoi":"10.21203/rs.3.rs-5435843/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5435843/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe world health organization (WHO) has officially declared the COVID-19 as a global pandemic in March 2020. Consequently, nations worldwide took some preventive measures, including lockdowns, quarantines, and social distancing to slow down the spread of coronavirus. This unprecedented event has profoundly disrupted the normal way of life. The pandemic had devastating impacts on various aspects of society such as healthcare systems, social life, the economy, and education. People from around the world began expressing emotions of fear, isolation, and various kinds of traumatic disorders on social media networks such as Twitter and Facebook. This research paper explores the impacts of COVID-19 in Morocco using topic modeling, sentiment analysis, and time series analysis. The study follows a two-step process. Initially, we employed a topic model, specifically BERTopic, to extract the main themes from a dataset containing comments gathered from the online newspaper Hespress and Twitter. Subsequently, we conducted a topic-based sentiment analysis to assess how COVID-19 has impacted Moroccans through a time window of three years. The findings revealed that sentiments related to the various topics were highly negative. In addition, we leveraged time-series data on COVID-19 to examine how the evolving epidemiological situation influenced sentiments from March 2020, the beginning of the pandemic, until the end of 2022. Our analysis indicated a strong correlation between changes in COVID-19 cases and sentiment analysis results.\u003c/p\u003e","manuscriptTitle":"Investigating the various impacts of COVID-19 using Sentiment Analysis and Topic Modeling over three years","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-11-13 06:50:22","doi":"10.21203/rs.3.rs-5435843/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"0cc25be4-85fd-46bd-b165-4de99b4f2d27","owner":[],"postedDate":"November 13th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-11-13T06:50:22+00:00","versionOfRecord":[],"versionCreatedAt":"2024-11-13 06:50:22","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5435843","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5435843","identity":"rs-5435843","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}