The Future of Artificial Intelligence: Evaluating ChatGPT's Performance in X Sentiment Prediction.

The Future of Artificial Intelligence: Evaluating ChatGPT's Performance in X Sentiment Prediction. | 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 The Future of Artificial Intelligence: Evaluating ChatGPT's Performance in X Sentiment Prediction. Laila Malas, Ahmad Shawaqfeh, Ahmad AbuShakra This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5180421/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 18 Dec, 2024 Read the published version in Discover Artificial Intelligence → Version 1 posted 4 You are reading this latest preprint version Abstract Technological advancements have significantly progressed from the inception of computers and the internet to the rise of artificial intelligence (AI), profoundly impacting various sectors such as business, education, and healthcare. Among these advancements, ChatGPT has emerged as a prominent conversational AI tool, facilitating tasks such as sentiment prediction and sector-specific decision-making. This study aims to evaluate ChatGPT's performance in analyzing public sentiment using data from X (formerly Twitter) and to propose an ethical framework for its sector-specific applications. By combining the BERT transformer model with BiLSTM, Random Forest, and K-Nearest Neighbor algorithms, the study achieves a hybrid approach to sentiment prediction, demonstrating superior performance with an accuracy of 86.7%. Furthermore, the study explores ethical considerations, such as fairness, accountability, and transparency, to address ChatGPT's adoption across industries. The findings offer insights into both the technical and ethical dimensions of ChatGPT’s integration into business, education, and healthcare sectors, emphasizing its potential for sustainable and responsible use. Artificial Intelligence ChatGPT Sentiment Analysis Prediction Sustainability 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 Figure 13 1 Introduction With constant advancements in technology, the world has experienced major transforma- tions, such as the first Industrial Revolution Mechanization and Productivity for materials and equipment’s that highlight workforce for labor to work harder and gain lower salaries, the emergence of calculators to be used in math in order to solve mathematical problems to the emergence of computers and the Internet to reconnect throughout the world society, and finally, the development of artificial intelligence that called 4 th Industrial revolution and up that emphasis on Machine Learning, Cloud Computing, Blockchain Technology and much more. These advancements have led to a decline in the job opportunities available to workers, and businesses have started relying on artificial intelligence to per- form tasks that were previously carried out by humans. By using drones and robots to do these jobs. This trend is expected to continue, with more automated tasks and businesses becoming increasingly to rely on technology to enhance descion making and strategic planning, transform customer experience and accelerate sustainability and technological innovations. Thus, it is imperative to incorporate these systems into ethical frameworks and develop appropriate policies to address them with Knowing human roles. This is especially true in light of the emergence of ChatGPT the beginning of 5th industrial revolution, refers to the combination of powerful AI systems and human-centric techniques that prioritize collaboration between humans and robots. Unlike the Fourth Industrial Revolution, which concentrated on automation, the Fifth emphasizes supplementing human capabilities with AI, as demonstrated by ChatGPT. which is widely recognized as the leading large-language conversation generative AI model developed by OpenAI. comprehending texts. This artificial intelligence system has demonstrated exceptional proficiency in processing natural languages, analyzing and deducing directions to make informed decisions. Consequently, there has been significant discussion among academic and technological circles. Its wide-ranging applications across various industries, including education to enhance the educational process, marketing to develop or refine products and services, software engineering to identify and rectify program vulnerabilities, and healthcare to implement the necessary health measures for patients, have further underscored its importance‎ [ 1 ] ChatGPT has sparked discussions on the future of business and the potential replace- ment of programmers, researchers, and analysts. Professor Northeastern University em- phasizes the unstoppable trend of AI’s ubiquity ‎[ 2 ], urging coexistence and adaptation. Public opinion on social media can help refine algorithms, address devel- opmental issues, rectify programming errors, and enhance efficiency in programming, scientific data, literature, and security. Using a survey undertaken in the United States involving individuals aged 18 to 65 years, approximately 75% of respondents unfamiliar with ChatGPT expressed their belief that it would become a trend. Among the individuals who were acquainted with this technology but had not yet utilized it, 41% indicated their perception of it as a potential trend. However, among those who had prior experience using the software, merely a third held the view that it would gain popularity temporarily.‎[ 3 ] Therefore, ChatGPT can understand and generate human-like text, which has enabled various sectors to benefit from it, such as text analysis, personal task automation, per- sonalized learning experiences, content creation, customer service, code generation, and debugging. Therefore, these systems must be treated with caution, as they can raise im- portant ethical concerns, such as the possibility of ChatGPT providing misleading or dis- turbing information to those who mislead, leading to people not trusting ChatGPT. More- over, its ability to generate human-like texts will lead to copyright infringement, content modification, and putting itself at risk as the ultimate knowledge authority.‎[ 4 ]‎[ 5 ] Therefore, this study aimed to anticipate and analyze people’s opinions about ChatGPT using the X Platform to extract the data. Thus, we can identify weaknesses, develop them, and place them within the ethical foundations and policies that must be put in place so that we can put them on the right path and use them in a proper manner without bias or errors. This will benefit the business sector and society by using them within the correct principles based on each company’s policy, usage, and the power to use these systems in a sustainable and innovative manner. This paper is organized and Structured as follow beginning by studying previous related papers on text data analysis and future projections, followed by an analysis of actual tweet data related to the ChatGPT. Then introduces a novel hybrid model that integrates BERT embeddings with BiLSTM, Random Forest, and K-Nearest Neighbor algorithms to enhance real-time sentiment prediction on social media platforms, such as X. In contrast to previous studies that employed conventional models, such as Naive Bayes or Support Vector Machines, which often proved inadequate in capturing nuanced, context-dependent sentiment, our approach utilizes the sequential and bidirectional nature of BiLSTM to analyze complex language patterns effectively. By incorporating BERT's advanced contextual embeddings, this model captures the dynamic and informal language prevalent on platforms such as X, resulting in a more accurate and unbiased sentiment analysis. This combination not only improves prediction accuracy, but also advances real-time social media analysis by addressing challenges in processing informal and rapidly evolving texts. The paper also will discuss ChatGPT’s strengths, weaknesses, prospects, competitors’ perspectives and political ethical frame- work needed to be addressed among ChatGPT with further details provided in subsequent sections. 2 Literature Review Considering the objective of this paper, this section will move through previous studies that reflect the evolution of artificial intelligence and its ramifications on various aspects of life and practical applications. The purpose of this examination was to assess ChatGPT’s competence in specific domains, particularly when it comes to sentiment analysis and its predictive capabilities. By addressing research gaps that have not been adequately explored in prior studies, this article aims to provide a comprehensive evaluation of the potential of ChatGPT and its impact on the future of artificial intelligence. In this study " The Present and Future of Artificial Intelligence ," ‎[ 6 ] they investigate the current landscape and future potential of AI. Their findings revealed that AI has made substantial inroads into everyday life, taking on tasks traditionally performed by humans across various industries, including commerce, medical care, and logistics. The researchers highlighted AI's diverse capabilities, such as identifying patterns, interpreting speech, and implementing automated processes in different sectors to cut labor expenses and improve task precision. Despite ongoing technological progress, the authors recognized that AI still has room for improvement. Looking ahead, the study predicted increased automation and opposition to efficiency enhancements, improved problem-solving abilities based on learned experiences, and the development of principled frameworks to leverage AI's potential while addressing associated risks. This strategy aims to create value for stakeholders by emphasizing the responsible application of AI technology. The perspective presented in the previous paper aligns with the assertions made in ‎[ 7 ] " The Future of AI: Transforming Tomorrow " regarding the potential of artificial intelligence to enhance performance across various sectors. In the healthcare domain, AI is expected to improve diagnostic accuracy and facilitate the development of personalized treatment plans, thereby enhancing patient care. Similarly, in the education sector, AI technologies are anticipated to evaluate student performance more effectively, identify specific areas for improvement, and provide relevant educational resources to optimize the learning process. The transportation sector is likely to benefit from the implementation of autonomous vehicles, which are projected to improve road safety and reduce accident rates. Furthermore, the paper emphasizes the importance of mitigating bias in AI systems and establishing comprehensive policies for their governance. Additionally, it underscores the significance of advancing research and technical capabilities to fully realize the potential of artificial intelligence and ensure its responsible development and deployment. In contrast, the paper ‎[ 8 ] “ Integration of natural language processing with artificial intelligence: a review paper " argues that the integration of natural language processing and artificial intelligence is crucial for the future of AI. This fusion is based on the idea of humanoid robots and improving human-machine interaction through the ability to understand and process natural human language in various dialects. Additionally, it involves creating chatbots and automating customer service systems to handle queries, offer suggestions, and engage in sophisticated conversations to ensure client satisfaction as a practical application. This viewpoint sparks a lively debate about the direction of artificial intelligence research and its potential impact on the future. From the perspective of natural language processing in the context of artificial intelligence's future development, it is evident that virtual assistant tools have become essential for enhancing business performance. Consequently, it is necessary to evaluate one of these tools, specifically ChatGPT, which has gained widespread usage during this period. The research paper " Performance of ChatGPT on Chinese Master's Degree Entrance Examination in Clinical Medicine " ‎[ 9 ] examined the entrance examinations for a master's degree in clinical medicine in China to assess the reliability and performance of ChatGPT. The study utilized 300 samples of multiple-choice questions from 2021 and 2022, maintaining the original context without modifications or the addition of images. The findings revealed that ChatGPT achieved the highest scores in the field of medical humanities at 93.75%, while ChatGPT 3.5 and ChatGPT 4 obtained the lowest scores in the Department of Pathology at 37.5% and in the Department of Biochemistry at 60.23%, respectively. These results indicate ChatGPT's proficiency in medical knowledge; however, they also suggest the need for supervision, development, and rigorous evaluation, accompanied by proactive measures to address the prevailing limitations in specific specializations and the nature of questions it can comprehend. However, the study ‎[ 10 ] “ Evaluating the Performance of ChatGPT for Spam Email Detection " addressed the significance of email communication and the growing challenge of spam, which has undermined user trust and usability. To address this issue, spam detection is crucial for risk mitigation. The researchers aimed to assess the effectiveness of ChatGPT in identifying spam using both English and Chinese datasets. They employed a context-based learning approach to provide a model with instructional examples to gauge its spam-detection capabilities. The study compared the performance of ChatGPT with conventional methods, including Naive Bayes, Support Vector Machines (SVM), Logistic Regression (LR), and Dense Neural Networks. The results indicated that the spam detection performance of ChatGPT was inferior to that of traditional deep learning models, particularly when analyzing large English datasets. However, its performance improved with Chinese-language data because of the context-limited nature of the task. This suggests that the efficacy of context-based learning may vary depending on the extent of guidance provided to the model. To connect natural language processing with sentiment analysis Perceptions to guide the future development of artificial intelligence this paper, “ Mining Twitter for Insights into ChatGPT Sentiment: A Machine Learning Approach ," ‎[ 11 ] addresses the gap in understanding public perception of ChatGPT, a technology used for answering questions and generating text. By collecting Twitter data containing user opinions, the researchers categorized sentiments as positive, negative, or neutral. These categorized data were then input into machine learning algorithms, including logistic regression and support vector machines, to analyze public sentiment. The results revealed that the majority of people held neutral feelings towards ChatGPT, with a small percentage expressing either positive or negative sentiments. This finding underscores the importance of considering trust, transparency, and ethical implications when utilizing ChatGPT technology. Another research titled “ Sentiment Analysis in ChatGPT Interactions: Unraveling Emotional Dynamics, Model Evaluation, and User Engagement Insights ” ‎[ 12 ] examined how conversational AI systems like ChatGPT engage with users through text-based exchanges. The study emphasized the significance of sentiment analysis in accurately gauging user emotions and assessing the model's ability to interpret user-expressed sentiment. The researchers gathered data from ChatGPT interactions, conducted analyses, tested hypotheses to evaluate results, and input findings into the Naive Bayes Classifier model. The analysis demonstrated that user inquiries tend to be more positive than ChatGPT responses, and the Naive Bayes Classifier model demonstrated robust performance. However, this analysis has limitations, including potential biases in sentiment interpretation due to mixed data, restricted use of algorithms and evaluation methods, and an emphasis on qualitative aspects while overlooking quantitative elements in sentiment analysis. These shortcomings prompted further investigation into this topic. Previous research in sentiment analysis demonstrates limitations when applied to real-time, informal social media data, such as that from X, owing to the unique challenges presented by its dynamic and non-standardized language. Traditional models, including Naive Bayes and logistic regression, tend to exhibit inadequate performance with regard to contextual subtleties and the rapid evolution of social media language, frequently resulting in limited accuracy and generalizability. Additionally, a significant gap exists in the ethical evaluation of AI models for sentiment prediction, particularly concerning bias mitigation and sector-specific ethical standards. This study addresses these gaps by employing a hybrid BERT and BiLSTM model specifically designed to handle informal and complex linguistic structures, ensuring more reliable sentiment predictions. Furthermore, we integrate an ethical framework that evaluates the model's adherence to fairness, accountability, and transparency standards, thereby advancing sentiment analysis practices while promoting responsible AI application across diverse sectors. 3 Methodology To validate the objective of this research study in predicting individuals' attitudes toward ChatGPT without bias, quantitative analysis methods were employed to examine data encompassing the sentiments of early adopters’ users toward ChatGPT. These data were extracted from X (formerly Twitter), a platform that targets diverse demographic and geographic categories, enabling comprehensive and clear sentiment analysis due to user interactions. Subsequently, to process this data, it was incorporated into the data life cycle, which aims to collect, clean, and predict sentiments using machine learning techniques. The selection of sentiment analysis was based on its ability to utilize extensive, real-time data from X, which offers a wide array of spontaneous public opinions. This method encompasses a broad spectrum of sentiments from various demographic groups, making it particularly effective for assessing ChatGPT's capabilities. Although surveys and expert evaluations provide structured input, they lack the immediacy and variety found in social media content. By incorporating sentiment analysis, a thorough and context-sensitive evaluation of ChatGPT's performance is ensured. 3.1 Data Collection and Description To start the process of Analyzing ChatGPT Sentiments, this paper begins by determining the sources of data and describing major aspects of it by moving through the following rigorous process to ensure the reliability and validity of the findings: 1. Data Source : This study utilized data from X (previously known as Twitter), a social media platform recognized for its lively and casual communication style, making it an excellent source for examining real-time public opinion. The data collection took place in January 2023, a time of increased attention on ChatGPT following its release. This specific period enabled the researchers to observe the initial adoption stage and the changing user perspectives. The dataset was acquired from Kaggle, a reputable source for high-quality datasets, which ensured the data's credibility and pertinence to the study. 2. Data Description : The dataset comprises 50,001 records with 20 attributes, including textual data (e.g., tweet content), metadata (e.g., language, user engagement metrics like replies, likes, and retweets), and hashtags. This diverse structure provided a multi-dimensional view of ChatGPT-related sentiment, allowing for a comprehensive analysis of user engagement and sentiment trends. 3. Dataset Actions and Steps: To enhance data quality and relevance, initial preprocessing steps were implemented. These procedures involved eliminating duplicate tweets, removing non-English content, and discarding tweets with insufficient text (such as one-word posts). This refinement process reduced irrelevant information and ensured that the dataset comprised substantial content suitable for sentiment analysis. 3.2 Data Cleaning and Preprocessing Data preprocessing is a crucial step in data mining, enhancing its effectiveness. The methods used during preprocessing significantly impact the results of analytical algorithms. Explanatory data analyses (EDA) are applied to explain features using Matplotlib Library in Python and sentiment analysis for tweets. The graphs conclude that: Graph 1 : The data shows that user translatio_ja is the leading user in discussing ChatGPT topics, with 60 tweets, indicating a greater interest in discussing ChatGPT topics. Graph 2 : The study reveals that around 36736 unique X replies primarily focus on ChatGPT, indicating the value of customer insights in optimizing strategies for optimal results. Graph 3: The ChatGPT software significantly enhances brand awareness, identity, image, and personality, with the highest retweet count of 42416, suggesting that users' mention can positively or negatively impact the brand. Graph 4 : The study reveals that ChatGPT users' highest number of likes on related tweets is 27141, indicating that increased likes boost emotions towards the platform, regardless of text positivity or negativity. Graph 5 : The graph shows the top user's tweets about ChatGPT on X, with an anticipated 47665 quotes, providing a quick overview of response activity and reactions. Graph 6 : The ChatGPT topic displays the most languages in dataset tweets, with English, Japanese, and Spanish languages ranking first in terms of popularity. Graph 7: The graph shows that the highest hashtag count for a single tweet is around 36414, indicating the importance of hashtag keywords in reaching users to the ChatGPT concept. After All these graphs are displayed The Word Cloud Natural Language Processing algorithm visualizes textual information, highlighting critical data by highlighting individual words' prominence, often used for social media platform analysis. The nltk library is imported using Python for sentiment analysis, using Sentiment Intensity Analyzer to determine sentiment for each tweet based on polarity score (0-100). A score less than zero indicates negative sentiment, while a score over zero indicates positive sentiment. Neutral sentiment is indicated. The Word Cloud Algorithm is utilized after generating sentiments to identify negative, positive, and neutral word sentiments. According to the algothim used to test polarity and apply word Cloud technique to visualize it the result shown that: 1. ChatGPT, an AI platform developed by OpenAI, offers instant answers to various questions, unlike Google's list of websites. It also assists businesses in resolving problems, such as customer service chatbots, which are expected to become the future of jobs. Positive sentiments include AI, OpenAI, and the future of jobs. 2. Microsoft has announced a multibillion-dollar investment in ChatGPT, making it the exclusive provider of cloud computing service to OpenAI, resulting in neutral sentiment words. 3. The study reveals that ChatGPT can potentially destroy human intelligence if people become overly dependent on it, leading to cheating in exams and inaccuracies in solving mathematical and coding problems. Additionally, ChatGPT fails to provide accurate citations for references not mentioned on the internet or journals. The dataset stores sentiments as text, creating a new data frame called Sentiment.csv for prediction. This frame undergoes text preprocessing to prevent errors and bias, using a preprocessing step as illustrated in Fig. 11 Error! Reference source not found. Step 1 : Using the Text Sentiments values for Label encoding that converts categorical values into numerical values where this dataset assigns each value to an integer number in these sentiments the positive is assigned as 1, the Negative as -1, and the Neutral as 0. Step 2: Lowercasing The entirety of the text underwent conversion to lowercase characters to maintain consistency and mitigate case sensitivity issues. This step proved crucial in addressing the varied linguistic patterns and irregular capitalization commonly observed in X data. Step 3 : Tokenization: The dataset underwent tokenization to segment tweets into individual lexical units. This process facilitated a more granular sentiment analysis, enabling the model to comprehend the function of each word within the context of a sentence. Step 4: Stopword Removal: Common stopwords (e.g., "and," "the") were eliminated to mitigate noise and emphasize sentiment-bearing words, thereby enhancing computational efficiency and relevance. Step 5: Remove nonletters in tweet texts by removing non-alphabetical letters by comparing the character using ASCII value and checking if the character is not in that range will be removed from the text by Python, we can apply this process using functions that check if this character from ASCII range or not otherwise it will erase the letter. Step 6 : BERT Embeddings BERT was utilized to generate contextual embeddings for each tweet, capturing nuanced meanings and relationships between words. BERT's capacity to process word context within a sentence rendered it highly effective for analyzing informal and ambiguous language on X. Step 7: Apply stemming which is the process that reduces a word from its suffix, prefix, and affixes and apply it as a basic form such as waiting became wait after stemming, this paper uses Porter stemmer which is one of the most commonly used algorithms for stemming and it is based on simple rules and enhance model recognition Step 8 : Determine the target variable and set it alone so that it is understood more clearly which the text sentiment after label encoding and you want to gain a deeper understanding of it. Step 9 : after that text Sentiments will be converted to numerical values after splitting, stemming, and removing nonletters all text will be vectorized, and reduce number of errors as much as possible by applying optimization to get the best parameters for machine learning models. Step 10 : apply train, test split for the dataset which is the most important aspect in machine learning so we can test hypotheses quickly and inexpensively which divide the training data to perform it on the model in the first half and test the hypotheses for improvement and evaluation on the other half. This dataset has been divided into 80% training and 20% Testing for evaluation. Now the Textual data is cleaned it is ready to apply algorithms on it for modeling. 3.3 Data Modeling Using Hybrid Machine and Deep Learning Techniques To achieve robust and accurate sentiment predictions, we implemented a hybrid approach that combines BERT embeddings with three distinct algorithms: Random Forest, BiLSTM, and K-Nearest Neighbor (KNN). Each algorithm was selected based on its unique strengths and complementary capabilities in addressing sentiment analysis, particularly in the context of informal, dynamic social media data. 1.1.1 BERT Transformer with Random Forest Algorithm The selection of Random Forest as a baseline model was predicated on its capacity to effectively manage extensive datasets comprising diverse features. This ensemble technique constructs numerous decision trees during the training phase and synthesizes their outputs, thereby enhancing predictive accuracy and diminishing overfitting. Random Forest exhibits particular proficiency in handling datasets characterized by intricate and heterogeneous features, exemplified by those extracted from X, where sentiment indicators may display considerable variability across tweets. The incorporation of this model establishes a dependable benchmark for assessing the efficacy of more advanced analytical approaches. 1.1.2 BERT Transformer with Bidirectional Long -Short-Term Memory (LSTM) Model BiLSTM was selected for its superior capacity to process sequential and context-dependent data. In contrast to conventional algorithms, BiLSTM considers both forward and backward dependencies within a text, thereby capturing the comprehensive context of a tweet. This bidirectional approach is crucial for comprehending informal social media language, where word order and surrounding context significantly influence sentiment. Furthermore, BiLSTM utilizes deep learning to capture long-term dependencies, rendering it particularly suitable for sentiment analysis tasks that necessitate nuanced interpretation of language. Its integration with BERT embeddings further enhances its performance by providing rich contextual information at the word level. Applying the Following algothim Using Python as follow: The neural network model devised for text classification is characterized by a succession of intricate layers tailored to extract salient characteristics from textual data and assign it to one of three distinct categories. Commencing its architectural configuration is an embedding layer, tasked with transforming individual words into compact vectors with a consistent dimension. This pivotal stage aids the model in comprehending the semantic connotations of words by situating them within an expansive multidimensional context. Essential parameters including, but not limited to, vocab_size, embedding_size, and input_length delineate the fundamental attributes governing this layer's functionality. Subsequent to the embedding layer, a one-dimensional convolutional layer denoted as Conv1D is implemented. This layer orchestrates the application of assorted filters to the embedded sequences, thereby capturing localized patterns intrinsic to the textual corpus. The nuanced interplay among parameters such as filters, kernel_size, padding, and activation dictates the layer's operational paradigm. Transitioning further, a max-pooling layer recognized as MaxPooling1D is enlisted to curtail the dimensionality of the feature maps derived from the antecedent convolutional stratum. This tactical maneuver facilitates the extraction of paramount features embedded within the dataset. Consequent to this interim, a bidirectional LSTM (Long Short-Term Memory) layer is seamlessly integrated. LSTM stands as an exemplary archetype of a recurrent neural network distinguished by its proficiency in assimilating protracted dependencies within the textual expanse. Embracing a bidirectional orientation, this layer can adeptly navigate through input sequences in both a forward and reverse trajectory, thus capturing interdependencies extending across temporal dimensions. The parameter 32 is exclusively dedicated to specifying the dimensional complexity of the resultant output space. In a bid to fend off instances of overfitting, a judiciously positioned dropout layer is enshrined within the model's structural framework. Functioning akin to a stochastic gatekeeper, this layer indiscriminately rejects a fraction of input units throughout the training phase. In this particular scenario, a substantial 40% of input units are earmarked for potential exclusion. Finally, dense output of three units corresponds to the three classes in the classification job. In this case, the softmax activation function was used to derive probabilities for each class. The ultimate hidden states produced by the BiLSTM served as the input for a densely connected classification layer. This layer was designed to determine the sentiment of a tweet (either positive or negative) by leveraging the combined contextual understanding derived from BERT and sequential processing capabilities of BiLSTM. A model was built as described above. It was then compiled with a categorical cross-entropy loss function, which is fully appropriate for multi-class classification tasks. The Stochastic Gradient Descent optimizer was used. Metrics for model evaluation were specified to be based on accuracy, precision, and recall. 1.1.3 BERT Transformer with K-Nearest Neighbor K-nearest neighbors (KNN) was incorporated as a comparatively straightforward yet efficacious model for sentiment classification based on proximity to analogous data points in the embedding space. Its inherent simplicity renders it robust in managing noisy datasets, wherein subtle variations in language and structure might otherwise introduce errors. By utilizing BERT embeddings, KNN leverages a high-dimensional representation of text, enabling the identification of patterns and similarities that are crucial for accurate sentiment classification. Its function in this study is to provide a comparative perspective, elucidating the advantages of more sophisticated deep learning models such as BiLSTM while demonstrating its utility in scenarios where computational efficiency is paramount. Each model contributes distinct and complementary aspects to the sentiment analysis task. Random Forest provides a baseline of reliability, KNN offers computational efficiency, and BiLSTM facilitates advanced contextual understanding. Collectively, these models present a comprehensive assessment of sentiment prediction capabilities, with BERT embeddings serving as a unifying foundation that enhances the overall performance of all three algorithms. 1.2 Model Evaluation and Exploration The model's overall correctness is measured by accuracy, while precision and recall evaluate the balance between false positives and false negatives. By combining these two metrics, the F1-score offers a comprehensive assessment of the model's performance, which is particularly valuable when dealing with datasets that have uneven distributions of sentiment classes. Based on Table 1 Table 1 : ChatGPT Tweets Sentiment Prediction Algorithms Performance Algorithms Accuracy Precision Recall F1-Score BERT + Random Forest 84.3% 83.5% 85.0% 84.2% BERT + BiLSTM 86.7% 87.0% 86.5% 86.7% BERT + KNN 81.6% 80.5% 82.0% 81.2% In addition to accuracy, precision, recall, and F1-score were calculated to provide a thorough evaluation of model performance. For example, the BiLSTM with BERT model had an F1-score of 86.7%, which was nearly identical to its precision (87.0%) and recall (86.5%). These metrics demonstrate that the model performs a decent job of reducing false positives and negatives. Furthermore, the confusion matrix (Figure 12) demonstrates a considerable reduction in false negatives when compared to other models, confirming the BiLSTM's effectiveness in capturing nuanced sentiment in informal social media data. The findings highlight the vital role of BERT's contextual embeddings and BiLSTM's sequential processing in accurately predicting sentiment from social media content. These outcomes indicate that hybrid deep learning approaches outperform conventional machine learning techniques when dealing with the dynamic and informal language typical of platforms such as X. This superiority is particularly evident in handling the unique characteristics of social media text. 1.3 Data Visualization The dashboard outlines the performance of ChatGPT tweets, focusing on word cloud topics, user distribution, and trend over time. Key performance indicators (KPIs) are set to help users understand ChatGPT's limitations and advantages. Power BI is used to visually represent the data, providing a meaningful understanding of the chatbot's performance. Which is illustrated in Fig. 13 This dashboard gives us recommendations for users that ChatGPT, an AI tool, is used by students to perform tasks, but it can lead to lazy students. It functions as a Google search engine, but in a different way. The dashboard displays the number of tweets, with positive, negative, and neutral percentages. It's important to understand the limitations of ChatGPT and its potential impact on students' learning. 1.4 Ethical validation and Bias Mitigation The primary objective of this study was to ensure unbiased and ethical AI predictions given the potential societal and organizational implications of sentiment analysis. To address these concerns, the following ethical guidelines and validation procedures were incorporated: 1- Bias mitigation in data pre-processing: During preprocessing, procedures such as tokenization, stopword removal, and stemming were implemented to standardize text inputs and minimize linguistic biases. For instance, stopword removal ensured that irrelevant yet common words (e.g., "the," "and") did not disproportionately influence the sentiment analysis process. 2- Non-alphanumeric characters, emojis, and hashtags are systematically processed to retain their contextual sentiment contribution without introducing bias based on their frequency or representation. 3- Fair Representation of Sentiments. The dataset was examined to ensure that all the sentiment classes (positive, negative, and neutral) were adequately represented. This step mitigates the risk of model overfitting to a dominant sentiment class, which is a prevalent issue in real-world social media data. 4- Ethical Model Selection The models were selected on the basis of their capacity to provide explainable and fair predictions. BERT embeddings were chosen for their contextual richness, enabling more accurate comprehension of user sentiments across diverse linguistic and cultural backgrounds. The BiLSTM model was utilized to reduce sequential biases by considering both forward and backward contexts, ensuring fairness in interpreting informal and varied social media languages. 5- Fairness Audit and Evaluation training, the model predictions were evaluated across different demographic and linguistic subgroups to identify any potential biases in sentiment classification. This fairness audit facilitated refinement of the model parameters to ensure the equitable treatment of all data points. Confusion matrices were analyzed to detect disparities in model performance across sentiment classes, ensuring a balanced prediction accuracy. 6- Adherence to Ethical Standard: The study adhered to the principles of fairness, accountability, and transparency by employing unbiased preprocessing steps, rigorously evaluating model outputs, and documenting the decision-making process in detail. These measures align with globally recognized AI ethics guidelines and ensure the responsible application of AI technologies. 4 Results and Discussion Based on the methodology employed to analyze the performance and perceptions of individuals regarding ChatGPT, the results underscore the vital importance of contextual embedding (BERT) and sequential processing (BiLSTM) in obtaining high-precision sentiment analysis on dynamic social media platforms. This emphasizes the advantage of hybrid deep learning approaches over conventional methods when examining informal and rapidly changing texts, which is fundamental to platforms such as X. Although the BERT + BiLSTM model demonstrated impressive performance, it was not without limitations. The use of pre-established embeddings may not fully capture domain-specific subtleties or emerging trends on social media. Moreover, the computational demands and potential for overfitting of the BiLSTM model necessitate careful adjustment of parameters, particularly when applying the method to larger datasets. Subsequent studies should investigate ensemble techniques or specialized transformer architectures optimized for specific domains. Notwithstanding its superior performance, this investigation acknowledged that model accuracy constitutes only one component in the evaluation of machine learning systems. While hybrid deep learning method demonstrated superior efficacy in sentiment prediction, additional factors such as bias mitigation, ethical considerations, and implications of artificial intelligence deployment across various industries were subjected to critical examination based on the extant literature. Model Performance and Comparison with Previous Studies This study introduces a novel approach to sentiment analysis by employing the BiLSTM model with BERT Transformer, which represents an advancement over the previous research. While earlier investigations ‎[11‎,12] relied on logistic regression or machine learning algorithms, such as support vector machines and Naive Bayes, they failed to effectively capture long-term dependencies and contextual information in written data. The BiLSTM model with BERT (Hybrid method) utilized in this research offers a sophisticated method for analyzing sentiment in text, considering both forward and backward aspects. Previous studies acknowledged the issue of bias in sentiment interpretation but did not provide concrete solutions. To address this gap, our study implemented an effective bias mitigation strategy, including BERT Transformer, text normalization and stop word removal, resulting in more precise outcomes. Additionally, while past research has primarily focused on improving accuracy, it has often overlooked other ethical considerations. Our approach advances the field by incorporating performance enhancement vertices and improvement techniques for bias mitigation, making it not only efficient, but also ethically sound for developing responsible AI systems. Our research distinguishes itself from prior studies by emphasizing the significance of combining accuracy and context while using AI responsibly, thus addressing sentiment analysis from a more comprehensive perspective. Bias and Ethical Considerations The ethical framework was created using the principles of fairness, accountability, and transparency (FAT). Its development involves three essential steps: (1) identifying ethical constraints unique to AI in each area, (2) examining existing guidelines such as the OECD AI principles, and (3) adapting recommendations to address sector-specific concerns. For example, in healthcare, the framework prioritizes patient data protection, whereas in education, it concentrates on eliminating plagiarism and ensuring equal access. These concepts were used to assess ChatGPT's compliance with ethical norms in the business, education, and healthcare sectors. In certain studies, ‎[ 7 ] shows the significance of addressing bias in artificial intelligence (AI) systems has been emphasized in the domains of healthcare and education, where inaccurate predictions can have substantial real-world implications because they are sensitive. With the incorporation of text processing into algorithms such as BILSTM with BERT, there has been a concerted effort to mitigate the bias from model predictions. Through the analysis and processing of texts to eliminate words that do not influence sentiment, this study aims to ensure that the model's predictions are not disproportionately affected by specific phrases or linguistic patterns. This approach is crucial for reducing the biases that might otherwise emerge from the linguistic or demographic characteristics of the dataset. However, although these measures contribute to fairness, they do not eliminate the risk of bias entirely. Ongoing investigations should focus on developing additional techniques to detect and address bias, ensuring that artificial intelligence systems deliver fair results for all individuals who use them. Subsequently, based on these results, it is imperative to elucidate the negative aspects of ChatGPT, necessitating an understanding of the ethical implications associated with its use and potential mitigation strategies. Privacy concerns arise when individuals share personal information with ChatGPT for advice, as the AI model may inadvertently disclose sensitive details from the user's past, resulting in a breach of privacy. Furthermore, organizations may utilize ChatGPT to engage with customers and obtain their information without explicit consent, constituting an unethical misuse of data and a violation of privacy. Additionally, ChatGPT can be inappropriately employed to fabricate or disseminate information with the intent of causing harm to individuals or exploiting the technology for academic misconduct in scientific and literary fields, such as publishing plagiarized research papers, thereby compromising integrity and intellectual honesty. The above-mentioned ethical implications necessitate the implementation of strategies to address them when utilizing the ChatGPT. These strategies include establishing policies for AI developers to provide transparent information regarding data usage, collection, and storage; implementing systems that facilitate the auditing and tracking of AI decisions; and ensuring accountability and transparency. Furthermore, it is essential to enhance training data to mitigate biases and ensure integrity as well as to create mechanisms for users to report potentially harmful or unethical AI behaviors. Of paramount importance is the incorporation of human oversight as a fundamental component in reviewing and intervening in AI decision-making processes to prevent ethical concerns that may affect efficiency and productivity in the workplace. ‎[ 14 ] Field-Specific Implications The specific implications of this study are extensive and have exerted a significant influence on numerous industries, including business, healthcare, finance, and education, which have experienced substantial impacts through the application of natural language processing and artificial intelligence models encompassing the development of sentiment analysis utilizing artificial intelligence by integrating advanced deep learning into text-analysis-based fields using BILSTM with BERT and enhancing prediction accuracy in sentiment analysis. This is beneficial in marketing, customer service, and social media analysis, enabling organizations to gain a more comprehensive understanding of consumer opinions and sentiments. In Addition to the mitigation of bias in artificial intelligence systems through the implementation of text-processing techniques, such as converting text to lowercase letters and removing superfluous words, has a particular impact on the healthcare, education, and financial services sectors, where biased predictions can have serious real-world consequences. Ensuring fairness in artificial intelligence models is crucial for providing equitable services across diverse demographics. However, the study acknowledges that these measures do not completely eliminate bias, indicating that continued research is necessary to further develop unbiased artificial intelligence systems. Lastly, ethical considerations in deploying AI by designating human resources as the primary decision-maker are of paramount importance. This is particularly critical in domains such as human resources, legal services, and public administration, where AI recommendations or decisions must be reviewed to prevent potential ethical or legal violations. Industries must establish robust oversight mechanisms to ensure that AI systems augment human decision-making, rather than supplant it in sensitive contexts. 5 Conclusion and Future Work recommendation This study aimed to analyze and predict individuals' sentiments towards ChatGPT without bias, examine its impact on various industries, and evaluate the ethical considerations that should be considered when utilizing ChatGPT, employing machine and deep learning algorithms in natural language processing. The results of this investigation demonstrated that the Bidirectional LSTM with BERT model achieved significant superiority in performance and bias mitigation compared to machine learning algorithms with BERT owing to its ability to analyze data in multiple directions, yielding more accurate results. The study also explored how ethical considerations can be incorporated when using ChatGPT and its primary impacts on several sectors, including education, healthcare, and business. Consequently, future research should focus on developing additional techniques to further reduce bias in sentiment analysis, while emphasizing the importance of fairness and accuracy in sentiment prediction. Moreover, further research should address the broader ethical implications of deploying these models in sensitive industries such as healthcare and finance, where AI-driven decisions can have far-reaching consequences. Declarations Ethical Consideration We declare that all experiments and procedures conducted in this research adhere to the ethical standards outlined by Princess Sumaya University for Technology. Informed consent was obtained from all participants involved in the study, and their anonymity and confidentiality were strictly maintained. Conflicts of Interest: The authors declare that they have no conflicts of interest that could influence the outcome or interpretation of the research presented in this paper. Funding: No funding was received to assist with the preparation of this manuscript. Author Contribution Statement of Author Contributions:The research methodology and conceptualization were spearheaded by Laila Malas, who also played a key role in drafting the introduction. Ahmad Shawaqfeh took charge of the literature review, offering crucial insights and scholarly context. The final review and refinement of the manuscript were handled by Ahmad Abushakra, who ensured its overall coherence, precision, and compliance with academic standards. The writing, editing, and final approval of the manuscript before submission involved the collaborative efforts of all authors. Acknowledgement The Corresponding author hereby to acknowledge their profound appreciation to all co-authors for their substantial contributions to this manuscript. The collective expertise and collaborative efforts of these individuals were instrumental in the completion of this research. Mr. Ahmad Shawaqfeh whose manage the literature review and provide valuable insights that enrich in depth of the discussion and Dr.Ahmad Abushakra who review and edit the manuscript enhancing the coherence of the final draft from the initial stages of idea development to the final edits, their steadfast dedication has been crucial in completing this project. We are grateful for the insightful feedback and conversations that helped us navigate obstacles and enhance our concepts. This publication exemplifies the effectiveness of collaboration and collective academic endeavor. The invaluable input from our colleagues was essential to our success in this endeavor. Data Availability Data is provided within the manuscript or supplementary information files” offer from Kaggel website under the following link : https://www.kaggle.com/datasets/tariqsays/chatgpt-twitter-dataset References Schönberger, M. (2023). ChatGPT in Higher Education: The Good, The Bad, and The University. Editorial Universitat Politècnica de València. 331-338. https://doi.org/10.4995/HEAd23.2023.16174 Kuzub, A. (2023, May 16). Should you be using ChatGPT? Experts say ‘yes,’ but don’t confuse it with a friend. Northeastern Global News. https://news.northeastern.edu/2023/05/16/using-chat-gpt/ VettaFi, LLC. (n.d.). VettaFi. VettaFi. https://insights.roboglobal.com/chatgpt-fact-from-fiction-expert-insights-%20vs.-public-opinion. Zhan, X., Xu, Y., & Sarkadi, S. (2023, July 19). Deceptive AI ecosystems: The case of ChatGPT. Proceedings of the 5th International Conference on Conversational User Interfaces, 35, 1–6. Presented at the CUI ’23: ACM conference on Conversational User Interfaces, Eindhoven Netherlands. https://doi.org/10.1145/3571884.3603754 Rivas, P., & Zhao, L. (2023). Marketing with ChatGPT: Navigating the ethical terrain of GPT-based chatbot technology. AI, 4(2), 375-384. https://doi.org/10.3390/ai4020019 Alieksieiev, M., & Kurenkov, V. (2024). The present and the future of artificial intelligence. In Collection of Scientific Papers «ΛΌГOΣ» (pp. 231–232). https://doi.org/10.36074/logos-24.05.2024.050 Geetha, V., Gomathy, C. K., Teja, R. S., & Aniketh, V. (2023). THE FUTURE OF AI: TRANSFORMING TOMORROW. INTERANTIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT, 07(11), 1–11. https://doi.org/10.55041/ijsrem27039 Chouhan, P., Paliwal, M., & Gupta, T. (2023). Integration of natural language processing with artificial intelligence: A review paper. Journal of Analysis and Computation, 17(2), 294–300. https://doi.org/ 10.30696/jac.xvii.2.2023.294-300 Li, K.-C., Bu, Z.-J., Shahjalal, M., He, B.-X., Zhuang, Z.-F., Li, C., … Liu, Z.-L. (2024). Performance of ChatGPT on Chinese master’s degree entrance examination in Clinical Medicine. PloS One, 19(4), e0301702. doi: 10.1371/journal.pone.0301702 Wu, Y., Si, S., Zhang, Y., Gu, J., & Wosik, J. (2024). Evaluating the performance of chatgpt for spam email detection. arXiv preprint arXiv:2402.15537. Sharma, S., Aggarwal, R., & Kumar, M. (2023, April). Mining Twitter for Insights into ChatGPT Sentiment: A Machine Learning Approach. In 2023 International Conference on Distributed Computing and Electrical Circuits and Electronics (ICDCECE) (pp. 1-6). IEEE. Esh, M. (2024). Sentiment Analysis in ChatGpt Interactions: Unraveling Emotional Dynamics, Model Evaluation, and User Engagement Insights. Technical Services Quarterly, 41(2), 160–174. https://doi.org/10.1080/07317131.2024.2319972 Tariq ·, M. (2AD), “ChatGPT Twitter Dataset”, Kaggle, Data set, Kaggle. OECD. (2024, May 3). OECD updates AI Principles to stay abreast of rapid technological developments. OECD. Retrieved from https://www.oecd.org/en/about/news/press-releases/2024/05/oecd-updates-ai-principles-to-stay-abreast-of-rapid-technological-developments.html. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 18 Dec, 2024 Read the published version in Discover Artificial Intelligence → Version 1 posted Editorial decision: Accepted 16 Dec, 2024 Submission checks completed at journal 11 Dec, 2024 Editor assigned by journal 04 Dec, 2024 First submitted to journal 15 Nov, 2024 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-5180421","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":389163723,"identity":"79eb019f-dbfb-448d-a963-37ce4e0c4b6d","order_by":0,"name":"Laila Malas","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA20lEQVRIiWNgGAWjYBACPhDxgIFBDspnBuID+LWwgYgEBgZjIMXYQJKWxAaEFgKAjYE78UNCTV36dv4z5g8YKqwTGxgP47eGjYF3s0TCscO5O2fkGDYwnEkHWncsgZCWDRIJbAdyN9zgMWxgbDsM1HLGgKAtPxL+1aUbnD8D1PKPOC3bJBLbmBMMDgAdxthAjBZm3m0WiX2HDXfOSCuckXAs3biNkF/42Xs33/jwrU7enP/whg8faqxl+yUIhBg8IsCOARnPJkFABxwg3M/fQKSWUTAKRsEoGCkAAGj9RxL17A9mAAAAAElFTkSuQmCC","orcid":"","institution":"Princess Sumaya University for Technology","correspondingAuthor":true,"prefix":"","firstName":"Laila","middleName":"","lastName":"Malas","suffix":""},{"id":389163724,"identity":"6ae9e70e-f980-42b0-8247-fcb4405cb43c","order_by":1,"name":"Ahmad Shawaqfeh","email":"","orcid":"","institution":"Princess Sumaya University for Technology","correspondingAuthor":false,"prefix":"","firstName":"Ahmad","middleName":"","lastName":"Shawaqfeh","suffix":""},{"id":389163725,"identity":"85f9feb6-1668-40e1-a083-224e211f367a","order_by":2,"name":"Ahmad AbuShakra","email":"","orcid":"","institution":"Princess Sumaya University for Technology","correspondingAuthor":false,"prefix":"","firstName":"Ahmad","middleName":"","lastName":"AbuShakra","suffix":""}],"badges":[],"createdAt":"2024-09-30 11:23:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5180421/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5180421/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s44163-024-00218-0","type":"published","date":"2024-12-18T15:57:42+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":71303662,"identity":"119a7c3a-5d66-456c-a474-fc20812a47a5","added_by":"auto","created_at":"2024-12-13 05:59:40","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":326273,"visible":true,"origin":"","legend":"\u003cp\u003eThe Figure Illustrates the Temporal Development Beginning with the 1st industrial revolution Mechanization, mass Production, Computer, Artificial Intelligence reaching to Cognitive Systems that able to think like humans\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/4a57b1c62228cccb9dbd7fe0.png"},{"id":71303634,"identity":"7ef24f77-e6e5-4faa-8010-b2740ca4e282","added_by":"auto","created_at":"2024-12-13 05:58:44","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":62400,"visible":true,"origin":"","legend":"\u003cp\u003eData Life Cycle applied for X Sentiment Data beginning with data Collection and Description, Cleaning and Preprocessing Sentiments, Modeling Using Machine Learning Algorithms for Prediction until model evaluation and interpretation\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/8ddbcf3865188f58a4419bbb.png"},{"id":71303219,"identity":"eb25e465-21a0-4e55-9c5b-9f9591ef8905","added_by":"auto","created_at":"2024-12-13 05:57:53","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":62059,"visible":true,"origin":"","legend":"\u003cp\u003eThe number of Tweets for each username that on his Tweet mention about ChatGPT\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/bda6f3e32b8911feb4238055.png"},{"id":71303772,"identity":"9c6aea12-f98e-4449-a0da-5950cc52bafb","added_by":"auto","created_at":"2024-12-13 06:01:20","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":34996,"visible":true,"origin":"","legend":"\u003cp\u003eillustrate the number of unique replies for tweet mention about ChatGPT\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/bb954c6c8d8b3430c5d02436.png"},{"id":71303218,"identity":"47114f3f-636b-4acc-b28e-a5caa7a7f354","added_by":"auto","created_at":"2024-12-13 05:57:53","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":24921,"visible":true,"origin":"","legend":"\u003cp\u003eThe number of retweets the users mention ChatGPT in their comments.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/ed59f2400c5bd91d6ae3290b.png"},{"id":71303637,"identity":"78b801aa-598e-4a2b-bf46-161f64d05150","added_by":"auto","created_at":"2024-12-13 05:58:46","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":31963,"visible":true,"origin":"","legend":"\u003cp\u003eThe number of likes for each ChatGPT tweet\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/134f22a35120edbd5c5c6810.png"},{"id":71303412,"identity":"bea386b9-c035-4736-ad79-402dbe7afebe","added_by":"auto","created_at":"2024-12-13 05:57:57","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":21903,"visible":true,"origin":"","legend":"\u003cp\u003ethe figure illustrates the number of quotes for each ChatGPT tweets\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/a83ae6ce8f6dceca75945532.png"},{"id":71303253,"identity":"71617672-71be-48ae-aa9f-3e4b93a95b9c","added_by":"auto","created_at":"2024-12-13 05:57:54","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":34107,"visible":true,"origin":"","legend":"\u003cp\u003eTop Languages That users mention more toward ChatGPT\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/4e23ce028e788f068de5292c.png"},{"id":71303903,"identity":"2f4a1f5d-d8d1-41eb-8d12-f99aba7e3255","added_by":"auto","created_at":"2024-12-13 06:05:52","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":45896,"visible":true,"origin":"","legend":"\u003cp\u003eThe Number of hashtags for each user tweet\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/ad6cd7d29cc7c7dcedb47349.png"},{"id":71303664,"identity":"2b5d629d-8247-4a24-9e4a-206390b532d7","added_by":"auto","created_at":"2024-12-13 05:59:45","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":675165,"visible":true,"origin":"","legend":"\u003cp\u003eSentiment Analysis Word Cloud most frequent terms (Positive, Neutral, and Negative) Sentiments\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/0942e97c77d5c08dede496d1.png"},{"id":71303642,"identity":"e5282daf-e821-459a-941a-1577d6c632e5","added_by":"auto","created_at":"2024-12-13 05:58:52","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":60424,"visible":true,"origin":"","legend":"\u003cp\u003eText Preprocessing and Cleaning Steps after Sentiments extraction to be ready for prediction\u003c/p\u003e","description":"","filename":"11.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/3a2bde22731433bba83f9263.png"},{"id":71303771,"identity":"7f9bea97-222d-44f1-99c7-c0f5b1e58adc","added_by":"auto","created_at":"2024-12-13 06:01:20","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":87561,"visible":true,"origin":"","legend":"\u003cp\u003eIllustrates the Confusion matrix for three algorithms, the first one for random forest, Secound one for Bidirectional LSTM model and third one for k nearest neighbor\u003c/p\u003e","description":"","filename":"12.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/62b6db9b29e910c73ec7684f.png"},{"id":71303964,"identity":"950a31c6-757e-4074-b18e-cdfb5d476cc6","added_by":"auto","created_at":"2024-12-13 06:07:44","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":404060,"visible":true,"origin":"","legend":"\u003cp\u003eSentiment Analysis Dashboard using Twitter ChatGPT showing key performance indicators in topic modeling, total tweets, Positive, negative and neutral sentiments, geographical distribution of using ChatGPT\u003c/p\u003e","description":"","filename":"13.png","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/522bce50f371ef907e6cb018.png"},{"id":72202734,"identity":"ed08ee86-a52c-47af-973d-0acb0a6d0f21","added_by":"auto","created_at":"2024-12-23 16:15:48","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2271654,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5180421/v1/ef6689aa-44d6-4119-9ac3-b70b08093d07.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"The Future of Artificial Intelligence: Evaluating ChatGPT's Performance in X Sentiment Prediction.","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eWith constant advancements in technology, the world has experienced major transforma- tions, such as the first Industrial Revolution Mechanization and Productivity for materials and equipment\u0026rsquo;s that highlight workforce for labor to work harder and gain lower salaries, the emergence of calculators to be used in math in order to solve mathematical problems to the emergence of computers and the Internet to reconnect throughout the world society, and finally, the development of artificial intelligence that called 4\u003cem\u003eth\u003c/em\u003e Industrial revolution and up that emphasis on Machine Learning, Cloud Computing, Blockchain Technology and much more. These advancements have led to a decline in the job opportunities available to workers, and businesses have started relying on artificial intelligence to per- form tasks that were previously carried out by humans. By using drones and robots to do these jobs. This trend is expected to continue, with more automated tasks and businesses becoming increasingly to rely on technology to enhance descion making and strategic planning, transform customer experience and accelerate sustainability and technological innovations.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThus, it is imperative to incorporate these systems into ethical frameworks and develop appropriate policies to address them with Knowing human roles. This is especially true in light of the emergence of ChatGPT the beginning of 5th industrial revolution, refers to the combination of powerful AI systems and human-centric techniques that prioritize collaboration between\u003c/p\u003e \u003cp\u003ehumans and robots. Unlike the Fourth Industrial Revolution, which concentrated on automation, the Fifth emphasizes supplementing human capabilities with AI, as demonstrated by ChatGPT. which is widely recognized as the leading large-language conversation generative AI model developed by OpenAI. comprehending texts. This artificial intelligence system has demonstrated exceptional proficiency in processing natural languages, analyzing and deducing directions to make informed decisions. Consequently, there has been significant discussion among academic and technological circles. Its wide-ranging applications across various industries, including education to enhance the educational process, marketing to develop or refine products and services, software engineering to identify and rectify program vulnerabilities, and healthcare to implement the necessary health measures for patients, have further underscored its importance\u0026lrm; [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eChatGPT has sparked discussions on the future of business and the potential replace- ment of programmers, researchers, and analysts. Professor Northeastern University em- phasizes the unstoppable trend of AI\u0026rsquo;s ubiquity \u0026lrm;[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e], urging coexistence and adaptation. Public opinion on social media can help refine algorithms, address devel- opmental issues, rectify programming errors, and enhance efficiency in programming, scientific data, literature, and security.\u003c/p\u003e \u003cp\u003eUsing a survey undertaken in the United States involving individuals aged 18 to 65 years, approximately 75% of respondents unfamiliar with ChatGPT expressed their belief that it would become a trend. Among the individuals who were acquainted with this technology but had not yet utilized it, 41% indicated their perception of it as a potential trend. However, among those who had prior experience using the software, merely a third held the view that it would gain popularity temporarily.\u0026lrm;[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eTherefore, ChatGPT can understand and generate human-like text, which has enabled various sectors to benefit from it, such as text analysis, personal task automation, per- sonalized learning experiences, content creation, customer service, code generation, and debugging. Therefore, these systems must be treated with caution, as they can raise im- portant ethical concerns, such as the possibility of ChatGPT providing misleading or dis- turbing information to those who mislead, leading to people not trusting ChatGPT. More- over, its ability to generate human-like texts will lead to copyright infringement, content modification, and putting itself at risk as the ultimate knowledge authority.\u0026lrm;[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u0026lrm;[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eTherefore, this study aimed to anticipate and analyze people\u0026rsquo;s opinions about ChatGPT using the X Platform to extract the data. Thus, we can identify weaknesses, develop them, and place them within the ethical foundations and policies that must be put in place so that we can put them on the right path and use them in a proper manner without bias or errors. This will benefit the business sector and society by using them within the correct principles based on each company\u0026rsquo;s policy, usage, and the power to use these systems in a sustainable and innovative manner.\u003c/p\u003e \u003cp\u003eThis paper is organized and Structured as follow beginning by studying previous related papers on text data analysis and future projections, followed by an analysis of actual tweet data related to the ChatGPT. Then introduces a novel hybrid model that integrates BERT embeddings with BiLSTM, Random Forest, and K-Nearest Neighbor algorithms to enhance real-time sentiment prediction on social media platforms, such as X. In contrast to previous studies that employed conventional models, such as Naive Bayes or Support Vector Machines, which often proved inadequate in capturing nuanced, context-dependent sentiment, our approach utilizes the sequential and bidirectional nature of BiLSTM to analyze complex language patterns effectively. By incorporating BERT's advanced contextual embeddings, this model captures the dynamic and informal language prevalent on platforms such as X, resulting in a more accurate and unbiased sentiment analysis. This combination not only improves prediction accuracy, but also advances real-time social media analysis by addressing challenges in processing informal and rapidly evolving texts. The paper also will discuss ChatGPT\u0026rsquo;s strengths, weaknesses, prospects, competitors\u0026rsquo; perspectives and political ethical frame- work needed to be addressed among ChatGPT with further details provided in subsequent sections.\u003c/p\u003e"},{"header":"2 Literature Review","content":"\u003cp\u003eConsidering the objective of this paper, this section will move through previous studies that reflect the evolution of artificial intelligence and its ramifications on various aspects of life and practical applications. The purpose of this examination was to assess ChatGPT\u0026rsquo;s competence in specific domains, particularly when it comes to sentiment analysis and its predictive capabilities. By addressing research gaps that have not been adequately explored in prior studies, this article aims to provide a comprehensive evaluation of the potential of ChatGPT and its impact on the future of artificial intelligence.\u003c/p\u003e \u003cp\u003eIn this study \"\u003cem\u003eThe Present and Future of Artificial Intelligence\u003c/em\u003e,\" \u0026lrm;[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] they investigate the current landscape and future potential of AI. Their findings revealed that AI has made substantial inroads into everyday life, taking on tasks traditionally performed by humans across various industries, including commerce, medical care, and logistics. The researchers highlighted AI's diverse capabilities, such as identifying patterns, interpreting speech, and implementing automated processes in different sectors to cut labor expenses and improve task precision. Despite ongoing technological progress, the authors recognized that AI still\u003c/p\u003e \u003cp\u003ehas room for improvement. Looking ahead, the study predicted increased automation and opposition to efficiency enhancements, improved problem-solving abilities based on learned experiences, and the development of principled frameworks to leverage AI's potential while addressing associated risks. This strategy aims to create value for stakeholders by emphasizing the responsible application of AI technology.\u003c/p\u003e \u003cp\u003eThe perspective presented in the previous paper aligns with the assertions made in \u0026lrm;[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e] \"\u003cem\u003eThe Future of AI: Transforming Tomorrow\u003c/em\u003e\" regarding the potential of artificial intelligence to enhance performance across various sectors. In the healthcare domain, AI is expected to improve diagnostic accuracy and facilitate the development of personalized treatment plans, thereby enhancing patient\u003c/p\u003e \u003cp\u003ecare. Similarly, in the education sector, AI technologies are anticipated to evaluate student\u003c/p\u003e \u003cp\u003eperformance more effectively, identify specific areas for improvement, and provide relevant educational resources to optimize the learning process. The transportation sector is likely to benefit\u003c/p\u003e \u003cp\u003efrom the implementation of autonomous vehicles, which are projected to improve road safety and\u003c/p\u003e \u003cp\u003ereduce accident rates. Furthermore, the paper emphasizes the importance of mitigating bias in AI systems and establishing comprehensive policies for their governance. Additionally, it underscores the significance of advancing research and technical capabilities to fully realize the potential of artificial intelligence and ensure its responsible development and deployment.\u003c/p\u003e \u003cp\u003eIn contrast, the paper \u0026lrm;[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] \u0026ldquo;\u003cem\u003eIntegration of natural language processing with artificial intelligence: a review paper\u003c/em\u003e\" argues that the integration of natural language processing and artificial intelligence is crucial for the future of AI. This fusion is based on the idea of humanoid robots and improving human-machine interaction through the ability to understand and process natural human language in various dialects. Additionally, it involves creating chatbots and automating customer service systems to handle queries, offer suggestions, and engage in sophisticated conversations to ensure client satisfaction as a practical application. This viewpoint sparks a lively debate about the direction of artificial intelligence research and its potential impact on the future.\u003c/p\u003e \u003cp\u003eFrom the perspective of natural language processing in the context of artificial intelligence's future development, it is evident that virtual assistant tools have become essential for enhancing business performance. Consequently, it is necessary to evaluate one of these tools, specifically ChatGPT, which has gained widespread usage during this period. The research paper \"\u003cem\u003ePerformance of ChatGPT on Chinese Master's Degree Entrance Examination in Clinical Medicine\u003c/em\u003e\" \u0026lrm;[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] examined the entrance examinations for a master's degree in clinical medicine in China to assess the reliability and performance of ChatGPT. The study utilized 300 samples of multiple-choice questions from 2021 and 2022, maintaining the original context without modifications or the addition of images. The findings revealed that ChatGPT achieved the highest\u003c/p\u003e \u003cp\u003escores in the field of medical humanities at 93.75%, while ChatGPT 3.5 and ChatGPT 4 obtained the lowest scores in the Department of Pathology at 37.5% and in the Department of Biochemistry at 60.23%, respectively. These results indicate ChatGPT's proficiency in medical knowledge; however, they also suggest the need for supervision, development, and rigorous evaluation, accompanied by proactive measures to address the prevailing limitations in specific specializations and the nature of questions it can comprehend.\u003c/p\u003e \u003cp\u003eHowever, the study \u0026lrm;[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e] \u0026ldquo;\u003cem\u003eEvaluating the Performance of ChatGPT for Spam Email Detection\u003c/em\u003e\" addressed the significance of email communication and the growing challenge of spam, which has undermined user trust and usability. To address this issue, spam detection is crucial for risk mitigation. The researchers aimed to assess the effectiveness of ChatGPT in identifying spam\u003c/p\u003e \u003cp\u003eusing both English and Chinese datasets. They employed a context-based learning approach to provide a model with instructional examples to gauge its spam-detection capabilities. The study\u003c/p\u003e \u003cp\u003ecompared the performance of ChatGPT with conventional methods, including Naive Bayes, Support Vector Machines (SVM), Logistic Regression (LR), and Dense Neural Networks. The results indicated that the spam detection performance of ChatGPT was inferior to that of traditional deep learning models, particularly when analyzing large English datasets. However, its performance improved with Chinese-language data because of the context-limited nature of the task. This suggests that the efficacy of context-based learning may vary depending on the extent of guidance provided to the model.\u003c/p\u003e \u003cp\u003eTo connect natural language processing with sentiment analysis Perceptions to guide the future development of artificial intelligence this paper, \u0026ldquo;\u003cem\u003eMining Twitter for Insights into ChatGPT Sentiment: A Machine Learning Approach\u003c/em\u003e,\" \u0026lrm;[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] addresses the gap in understanding public perception of ChatGPT, a technology used for answering questions and generating text. By collecting Twitter data containing user opinions, the researchers categorized sentiments as positive, negative, or neutral. These categorized data were then input into machine learning algorithms, including logistic regression and support vector machines, to analyze public sentiment. The results revealed that the majority of people held neutral feelings towards ChatGPT, with a small percentage expressing either positive or negative sentiments. This finding underscores the importance of considering trust, transparency, and ethical implications when utilizing ChatGPT technology.\u003c/p\u003e \u003cp\u003eAnother research titled \u0026ldquo;\u003cem\u003eSentiment Analysis in ChatGPT Interactions: Unraveling Emotional Dynamics, Model Evaluation, and User Engagement Insights\u003c/em\u003e\u0026rdquo; \u0026lrm;[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eexamined how conversational AI systems like ChatGPT engage with users through text-based exchanges. The study emphasized the significance of sentiment analysis in accurately gauging user emotions and assessing the model's ability to interpret user-expressed sentiment. The researchers\u003c/p\u003e \u003cp\u003egathered data from ChatGPT interactions, conducted analyses, tested hypotheses to evaluate results, and input findings into the Naive Bayes Classifier model. The analysis demonstrated that user inquiries tend to be more positive than ChatGPT responses, and the Naive Bayes Classifier model demonstrated robust performance. However, this analysis has limitations, including potential biases\u003c/p\u003e \u003cp\u003ein sentiment interpretation due to mixed data, restricted use of algorithms and evaluation methods, and an emphasis on qualitative aspects while overlooking quantitative elements in sentiment analysis. These shortcomings prompted further investigation into this topic.\u003c/p\u003e \u003cp\u003ePrevious research in sentiment analysis demonstrates limitations when applied to real-time, informal social media data, such as that from X, owing to the unique challenges presented by its dynamic and non-standardized language. Traditional models, including Naive Bayes and logistic regression, tend to exhibit inadequate performance with regard to contextual subtleties and the rapid evolution of social media language, frequently resulting in limited accuracy and generalizability. Additionally, a significant gap exists in the ethical evaluation of AI models for sentiment prediction, particularly concerning bias mitigation and sector-specific ethical standards. This study addresses these gaps by employing a hybrid BERT and BiLSTM model specifically designed to handle informal and complex linguistic structures, ensuring more reliable sentiment predictions. Furthermore, we integrate an ethical framework that evaluates the model's adherence to fairness, accountability, and transparency standards, thereby advancing sentiment analysis practices while promoting responsible AI application across diverse sectors.\u003c/p\u003e"},{"header":"3 Methodology","content":"\u003cp\u003eTo validate the objective of this research study in predicting individuals\u0026apos; attitudes toward ChatGPT without bias, quantitative analysis methods were employed to examine data encompassing the sentiments of early adopters\u0026rsquo; users toward ChatGPT. These data were extracted from X (formerly Twitter), a platform that targets diverse demographic and geographic categories, enabling comprehensive and clear sentiment analysis due to user interactions. Subsequently, to process this data, it was incorporated into the data life cycle, which aims to collect, clean, and predict sentiments using machine learning techniques.\u003c/p\u003e\n\u003cp\u003eThe selection of sentiment analysis was based on its ability to utilize extensive, real-time data from X, which offers a wide array of spontaneous public opinions. This method encompasses a broad spectrum of sentiments from various demographic groups, making it particularly effective for assessing ChatGPT\u0026apos;s capabilities. Although surveys and expert evaluations provide structured input, they lack the immediacy and variety found in social media content. By incorporating sentiment analysis, a thorough and context-sensitive evaluation of ChatGPT\u0026apos;s performance is ensured.\u003c/p\u003e\n\u003ch2\u003e3.1 Data Collection and Description\u0026nbsp;\u003c/h2\u003e\n\u003cp\u003eTo start the process of Analyzing ChatGPT Sentiments, this paper begins by determining the sources of data and describing major aspects of it by moving through the following rigorous process to ensure the reliability and validity of the findings:\u003c/p\u003e\n\u003cp\u003e1. \u003cstrong\u003eData Source\u003c/strong\u003e: This study utilized data from X (previously known as Twitter), a social media platform recognized for its lively and casual communication style, making it an excellent source for examining real-time public opinion. The data collection took place in January 2023, a time of increased attention on ChatGPT following its release. This specific\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eperiod enabled the researchers to observe the initial adoption stage and the changing user perspectives. The dataset was acquired from Kaggle, a reputable source for high-quality datasets, which ensured the data\u0026apos;s credibility and pertinence to the study.\u003c/p\u003e\n\u003cp\u003e2. \u003cstrong\u003eData Description\u003c/strong\u003e: The dataset comprises 50,001 records with 20 attributes, including textual data (e.g., tweet content), metadata (e.g., language, user engagement metrics like replies, likes, and retweets), and hashtags. This diverse structure provided a multi-dimensional view of ChatGPT-related sentiment, allowing for a comprehensive analysis of user engagement and sentiment trends.\u003c/p\u003e\n\u003cp\u003e3. \u003cstrong\u003eDataset Actions and Steps: To\u003c/strong\u003e enhance data quality and relevance, initial preprocessing steps were implemented. These procedures involved eliminating duplicate tweets, removing non-English content, and discarding tweets with insufficient text (such as one-word posts). This refinement process reduced irrelevant information and ensured that the dataset comprised substantial content suitable for sentiment analysis.\u003c/p\u003e\n\u003ch2\u003e3.2 \u0026nbsp; \u0026nbsp; Data Cleaning and Preprocessing\u003c/h2\u003e\n\u003cp\u003eData preprocessing is a crucial step in data mining, enhancing its effectiveness. The methods used during preprocessing significantly impact the results of analytical algorithms. Explanatory data analyses (EDA) are applied to explain features using Matplotlib Library in Python and sentiment analysis for tweets. The graphs conclude that:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGraph 1\u003c/strong\u003e: The data shows that user translatio_ja is the leading user in discussing ChatGPT topics, with 60 tweets, indicating a greater interest in discussing ChatGPT topics.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGraph 2\u003c/strong\u003e: The study reveals that around 36736 unique X replies primarily focus on ChatGPT, indicating the value of customer insights in optimizing strategies for optimal results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGraph 3:\u003c/strong\u003e The ChatGPT software significantly enhances brand awareness, identity, image, and personality, with the highest retweet count of 42416, suggesting that users\u0026apos; mention can positively or negatively impact the brand.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGraph 4\u003c/strong\u003e: The study reveals that ChatGPT users\u0026apos; highest number of likes on related tweets is 27141, indicating that increased likes boost emotions towards the platform, regardless of text positivity or negativity.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGraph 5\u003c/strong\u003e: \u0026nbsp; The graph shows the top user\u0026apos;s tweets about ChatGPT on X, with an anticipated 47665 quotes, providing a quick overview of response activity and reactions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGraph 6\u003c/strong\u003e: \u0026nbsp; The ChatGPT topic displays the most languages in dataset tweets, with English, Japanese, and Spanish languages ranking first in terms of popularity.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGraph 7:\u003c/strong\u003e\u0026nbsp; \u0026nbsp;The graph shows that the highest hashtag count for a single tweet is around 36414, indicating the importance of hashtag keywords in reaching users to the ChatGPT concept.\u003c/p\u003e\n\u003cp\u003eAfter All these graphs are displayed The Word Cloud Natural Language Processing algorithm visualizes textual information, highlighting critical data by highlighting individual words\u0026apos; prominence, often used for social media platform analysis.\u003c/p\u003e\n\u003cp\u003eThe nltk library is imported using Python for sentiment analysis, using Sentiment Intensity Analyzer to determine sentiment for each tweet based on polarity score (0-100). A score less than zero indicates negative sentiment, while a score over zero indicates positive sentiment. Neutral sentiment is indicated. \u0026nbsp; \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe Word Cloud Algorithm is utilized after generating sentiments to identify negative, positive, and neutral word sentiments.\u003c/p\u003e\n\u003cp\u003eAccording to the algothim used to test polarity and apply word Cloud technique to visualize it the result shown that:\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e1. ChatGPT, an AI platform developed by OpenAI, offers instant answers to various questions, unlike Google\u0026apos;s list of websites. It also assists businesses in resolving problems, such as customer service chatbots, which are expected to become the future of jobs. Positive sentiments include AI, OpenAI, and the future of jobs.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e2. Microsoft has announced a multibillion-dollar investment in ChatGPT, making it the exclusive provider of cloud computing service to OpenAI, resulting in neutral sentiment words.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e3.\u0026nbsp;The study reveals that ChatGPT can potentially destroy human intelligence if people become overly dependent on it, leading to cheating in exams and inaccuracies in solving mathematical and coding problems. Additionally, ChatGPT fails to provide accurate citations for references not mentioned on the internet or journals.\u003cstrong\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe dataset stores sentiments as text, creating a new data frame called \u003cstrong\u003eSentiment.csv\u003c/strong\u003e for prediction. This frame undergoes text preprocessing to prevent errors and bias, using a preprocessing step as illustrated in \u003cstrong\u003e\u003cem\u003eFig. 11\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003eError! Reference source not found.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 1\u003c/strong\u003e: Using the Text Sentiments values for Label encoding that converts categorical values into numerical values where this dataset assigns each value to an integer number in these sentiments the positive is assigned as 1, the Negative as -1, and the Neutral as 0.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 2:\u003c/strong\u003e Lowercasing The entirety of the text underwent conversion to lowercase characters to maintain consistency and mitigate case sensitivity issues. This step proved crucial in addressing the varied linguistic patterns and irregular capitalization commonly observed in X data.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 3\u003c/strong\u003e: Tokenization: The dataset underwent tokenization to segment tweets into individual lexical units. This process facilitated a more granular sentiment analysis, enabling the model to comprehend the function of each word within the context of a sentence.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 4:\u003c/strong\u003e Stopword Removal: Common stopwords (e.g., \u0026quot;and,\u0026quot; \u0026quot;the\u0026quot;) were eliminated to mitigate noise and emphasize sentiment-bearing words, thereby enhancing computational efficiency and relevance.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 5:\u003c/strong\u003e Remove nonletters in tweet texts by removing non-alphabetical letters by comparing the character using ASCII value and checking if the character is not in that range will be removed from the text by Python, we can apply this process using functions that check if this character from ASCII range or not otherwise it will erase the letter.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 6\u003c/strong\u003e: BERT Embeddings BERT was utilized to generate contextual embeddings for each tweet, capturing nuanced meanings and relationships between words. BERT\u0026apos;s capacity to process word context within a sentence rendered it highly effective for analyzing informal and ambiguous language on X.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 7:\u003c/strong\u003e Apply stemming which is the process that reduces a word from its suffix, prefix, and affixes and apply it as a basic form such as waiting became wait after stemming, this paper uses Porter stemmer which is one of the most commonly used algorithms for stemming and it is based on simple rules and enhance model recognition\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 8\u003c/strong\u003e: Determine the target variable and set it alone so that it is understood more clearly which the text sentiment after label encoding and you want to gain a deeper understanding of it.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 9\u003c/strong\u003e: after that text Sentiments will be converted to numerical values after splitting, stemming, and removing nonletters all text will be vectorized, and reduce number of errors as much as possible by applying optimization to get the best parameters for machine learning models.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 10\u003c/strong\u003e: apply train, test split for the dataset which is the most important aspect in machine learning so we can test hypotheses quickly and inexpensively which divide the training data to perform it on the model in the first half and test the hypotheses for improvement and evaluation on the other half. This dataset has been divided into 80% training and 20% Testing for evaluation.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNow the Textual data is cleaned it is ready to apply algorithms on it for modeling.\u003c/p\u003e\n\u003ch2\u003e3.3 \u0026nbsp; \u0026nbsp; Data Modeling Using Hybrid Machine and Deep Learning Techniques \u0026nbsp;\u003c/h2\u003e\n\u003cp\u003eTo achieve robust and accurate sentiment predictions, we implemented a hybrid approach that combines BERT embeddings with three distinct algorithms: Random Forest, BiLSTM, and K-Nearest Neighbor (KNN). Each algorithm was selected based on its unique strengths and complementary capabilities in addressing sentiment analysis, particularly in the context of informal, dynamic social media data.\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003e1.1.1 \u0026nbsp; \u0026nbsp; \u0026nbsp; BERT Transformer with Random Forest Algorithm\u0026nbsp;\u003c/h3\u003e\n\u003cp\u003eThe selection of Random Forest as a baseline model was predicated on its capacity to effectively manage extensive datasets comprising diverse features. This ensemble technique constructs numerous decision trees during the training phase and synthesizes their outputs, thereby enhancing predictive accuracy and diminishing overfitting. Random Forest exhibits particular proficiency in handling datasets characterized by intricate and heterogeneous features, exemplified by those extracted from X, where sentiment indicators may display considerable variability across tweets. The incorporation of this model establishes a dependable benchmark for assessing the efficacy of more advanced analytical approaches.\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003e1.1.2 \u0026nbsp; \u0026nbsp; \u0026nbsp; BERT Transformer with Bidirectional Long -Short-Term Memory (LSTM) Model\u0026nbsp;\u003c/h3\u003e\n\u003cp\u003eBiLSTM was selected for its superior capacity to process sequential and context-dependent data. In contrast to conventional algorithms, BiLSTM considers both forward and backward dependencies within a text, thereby capturing the comprehensive context of a tweet. This bidirectional approach is crucial for comprehending informal social media language, where word order and surrounding context significantly influence sentiment. Furthermore, BiLSTM utilizes deep learning to capture long-term dependencies, rendering it particularly suitable for sentiment analysis tasks that necessitate nuanced interpretation of language. Its integration with BERT embeddings further enhances its performance by providing rich contextual information at the word level.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Applying the Following algothim Using Python as follow: The neural network model devised for text classification is characterized by a succession of intricate layers tailored to extract salient characteristics from textual data and assign it to one of three distinct categories.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Commencing its architectural configuration is an embedding layer, tasked with transforming individual words into compact vectors with a consistent dimension. This pivotal stage aids the model in comprehending the semantic connotations of words by situating them within an expansive\u0026nbsp;\u003c/p\u003e\n\u003cp\u003emultidimensional context. Essential parameters including, but not limited to, vocab_size, embedding_size, and input_length delineate the fundamental attributes governing this layer\u0026apos;s functionality.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Subsequent to the embedding layer, a one-dimensional convolutional layer denoted as Conv1D is implemented. This layer orchestrates the application of assorted filters to the embedded sequences, thereby capturing localized patterns intrinsic to the textual corpus. The nuanced interplay among parameters such as filters, kernel_size, padding, and activation dictates the layer\u0026apos;s operational paradigm.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Transitioning further, a max-pooling layer recognized as MaxPooling1D is enlisted to curtail the dimensionality of the feature maps derived from the antecedent convolutional stratum. This tactical maneuver facilitates the extraction of paramount features embedded within the dataset.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Consequent to this interim, a bidirectional LSTM (Long Short-Term Memory) layer is seamlessly integrated. LSTM stands as an exemplary archetype of a recurrent neural network distinguished by its proficiency in assimilating protracted dependencies within the textual expanse. Embracing a bidirectional orientation, this layer can adeptly navigate through input sequences in both a forward and reverse trajectory, thus capturing interdependencies extending across temporal dimensions.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe parameter 32 is exclusively dedicated to specifying the dimensional complexity of the resultant output space. \u0026nbsp; \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn a bid to fend off instances of overfitting, a judiciously positioned dropout layer is enshrined within the model\u0026apos;s structural framework. Functioning akin to a stochastic gatekeeper, this layer indiscriminately rejects a fraction of input units throughout the training phase. In this particular scenario, a substantial 40% of input units are earmarked for potential exclusion.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Finally, dense output of three units corresponds to the three classes in the classification job. In this case, the softmax activation function was used to derive probabilities for each class.\u003c/p\u003e\n\u003cp\u003eThe ultimate hidden states produced by the BiLSTM served as the input for a densely connected classification layer. This layer was designed to determine the sentiment of a tweet (either positive or negative) by leveraging the combined contextual understanding derived from BERT and sequential processing capabilities of BiLSTM.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;A model was built as described above. It was then compiled with a categorical cross-entropy loss function, which is fully appropriate for multi-class classification tasks. The Stochastic Gradient Descent optimizer was used. Metrics for model evaluation were specified to be based on accuracy, precision, and recall.\u003c/p\u003e\n\u003ch3\u003e1.1.3 \u0026nbsp; \u0026nbsp; \u0026nbsp; BERT Transformer with K-Nearest Neighbor\u0026nbsp;\u003c/h3\u003e\n\u003cp\u003eK-nearest neighbors (KNN) was incorporated as a comparatively straightforward yet efficacious model for sentiment classification based on proximity to analogous data points in the embedding space. Its inherent simplicity renders it robust in managing noisy datasets, wherein subtle variations in language and structure might otherwise introduce errors. By utilizing BERT embeddings, KNN leverages a high-dimensional representation of text, enabling the identification of patterns and similarities that are crucial for accurate sentiment classification. Its function in this study is to provide a comparative perspective, elucidating the advantages of more sophisticated deep learning models such as BiLSTM while demonstrating its utility in scenarios where computational efficiency is paramount.\u003c/p\u003e\n\u003cp\u003eEach model contributes distinct and complementary aspects to the sentiment analysis task. Random Forest provides a baseline of reliability, KNN offers computational efficiency, and BiLSTM facilitates advanced contextual understanding. Collectively, these models present a comprehensive assessment of sentiment prediction capabilities, with BERT embeddings serving as a unifying foundation that enhances the overall performance of all three algorithms.\u003c/p\u003e\n\u003ch2\u003e1.2 \u0026nbsp; \u0026nbsp; Model Evaluation and Exploration\u003c/h2\u003e\n\u003cp\u003eThe model\u0026apos;s overall correctness is measured by accuracy, while precision and recall evaluate the balance between false positives and false negatives. By combining these two metrics, the F1-score offers a comprehensive assessment of the model\u0026apos;s performance, which is particularly valuable when dealing with datasets that have uneven distributions of sentiment classes. Based on\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable \u003cem\u003e1\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e1\u003c/strong\u003e: ChatGPT Tweets Sentiment Prediction Algorithms Performance\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"541\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAlgorithms\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 121px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1-Score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003eBERT + Random Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003e84.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 121px;\"\u003e\n \u003cp\u003e83.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e85.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e84.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003eBERT + BiLSTM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003e86.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 121px;\"\u003e\n \u003cp\u003e87.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e86.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e86.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003eBERT + KNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003e81.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 121px;\"\u003e\n \u003cp\u003e80.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e82.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e81.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eIn addition to accuracy, precision, recall, and F1-score were calculated to provide a thorough evaluation of model performance. For example, the BiLSTM with BERT model had an F1-score of 86.7%, which was nearly identical to its precision (87.0%) and recall (86.5%). These metrics demonstrate that the model performs a decent job of reducing false positives and negatives. Furthermore, the confusion matrix (Figure 12) demonstrates a considerable reduction in false negatives when compared to other models, confirming the BiLSTM\u0026apos;s effectiveness in capturing nuanced sentiment in informal social media data.\u003c/p\u003e\n\u003cp\u003eThe findings highlight the vital role of BERT\u0026apos;s contextual embeddings and BiLSTM\u0026apos;s sequential processing in accurately predicting sentiment from social media content. These outcomes indicate \u0026nbsp;\u003c/p\u003e\n\u003cp\u003ethat hybrid deep learning approaches outperform conventional machine learning techniques when dealing with the dynamic and informal language typical of platforms such as X. This superiority is particularly evident in handling the unique characteristics of social media text.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003e1.3 \u0026nbsp; \u0026nbsp; Data Visualization\u0026nbsp;\u003c/h2\u003e\n\u003cp\u003eThe dashboard outlines the performance of ChatGPT tweets, focusing on word cloud topics, user distribution, and trend over time. Key performance indicators (KPIs) are set to help users understand ChatGPT\u0026apos;s limitations and advantages. Power BI is used to visually represent the data, providing a meaningful understanding of the chatbot\u0026apos;s performance. Which is illustrated in \u003cstrong\u003eFig. 13\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis dashboard gives us recommendations for users that ChatGPT, an AI tool, is used by students to perform tasks, but it can lead to lazy students. It functions as a Google search engine, but in a different way. The dashboard displays the number of tweets, with positive, negative, and neutral percentages. It\u0026apos;s important to understand the limitations of ChatGPT and its potential impact on students\u0026apos; learning.\u003c/p\u003e\n\u003ch2\u003e1.4 \u0026nbsp; \u0026nbsp; Ethical validation and Bias Mitigation\u0026nbsp;\u003c/h2\u003e\n\u003cp\u003eThe primary objective of this study was to ensure unbiased and ethical AI predictions given the potential societal and organizational implications of sentiment analysis. To address these concerns, the following ethical guidelines and validation procedures were incorporated:\u003c/p\u003e\n\u003cp\u003e1-\u0026nbsp; \u0026nbsp;\u0026nbsp;Bias mitigation in data pre-processing: During preprocessing, procedures such as tokenization, stopword removal, and stemming were implemented to standardize text inputs and minimize linguistic biases. For instance, stopword removal ensured that irrelevant yet common words (e.g., \u0026quot;the,\u0026quot; \u0026quot;and\u0026quot;) did not disproportionately influence the sentiment analysis process.\u003c/p\u003e\n\u003cp\u003e2-\u0026nbsp; \u0026nbsp;\u0026nbsp;Non-alphanumeric characters, emojis, and hashtags are systematically processed to retain their contextual sentiment contribution without introducing bias based on their frequency or representation.\u003c/p\u003e\n\u003cp\u003e3-\u0026nbsp; \u0026nbsp;\u0026nbsp;Fair Representation of Sentiments. The dataset was examined to ensure that all the sentiment classes (positive, negative, and neutral) were adequately represented. This step mitigates the risk of model overfitting to a dominant sentiment class, which is a prevalent issue in real-world social media data.\u003c/p\u003e\n\u003cp\u003e4-\u0026nbsp; \u0026nbsp;\u0026nbsp;Ethical Model Selection The models were selected on the basis of their capacity to provide explainable and fair predictions. BERT embeddings were chosen for their contextual richness, enabling more accurate comprehension of user sentiments across diverse linguistic and cultural backgrounds. The BiLSTM model was utilized to reduce sequential biases by considering both forward and backward contexts, ensuring fairness in interpreting informal and varied social media languages.\u003c/p\u003e\n\u003cp\u003e5- \u0026nbsp; \u0026nbsp;Fairness Audit and Evaluation training, the model predictions were evaluated across different demographic and linguistic subgroups to identify any potential biases in sentiment classification. This fairness audit facilitated refinement of the model parameters to ensure the equitable treatment of all data points. Confusion matrices were analyzed to detect disparities in model performance across sentiment classes, ensuring a balanced prediction accuracy.\u003c/p\u003e\n\u003cp\u003e6- \u0026nbsp; \u0026nbsp;Adherence to Ethical Standard: The study adhered to the principles of fairness, accountability, and transparency by employing unbiased preprocessing steps, rigorously evaluating model outputs, and documenting the decision-making process in detail. These measures align with globally recognized AI ethics guidelines and ensure the responsible application of AI technologies.\u003c/p\u003e"},{"header":"4 Results and Discussion","content":"\u003cp\u003eBased on the methodology employed to analyze the performance and perceptions of individuals regarding ChatGPT, the results underscore the vital importance of contextual embedding (BERT) and sequential processing (BiLSTM) in obtaining high-precision sentiment analysis on dynamic social media platforms. This emphasizes the advantage of hybrid deep learning approaches over conventional methods when examining informal and rapidly changing texts, which is fundamental to platforms such as X.\u003c/p\u003e \u003cp\u003eAlthough the BERT\u0026thinsp;+\u0026thinsp;BiLSTM model demonstrated impressive performance, it was not without limitations. The use of pre-established embeddings may not fully capture domain-specific subtleties or emerging trends on social media. Moreover, the computational demands and potential for overfitting of the BiLSTM model necessitate careful adjustment of parameters, particularly when applying the method to larger datasets. Subsequent studies should investigate ensemble techniques or specialized transformer architectures optimized for specific domains.\u003c/p\u003e \u003cp\u003eNotwithstanding its superior performance, this investigation acknowledged that model accuracy constitutes only one component in the evaluation of machine learning systems. While hybrid deep learning method demonstrated superior efficacy in sentiment prediction, additional factors such as bias mitigation, ethical considerations, and implications of artificial intelligence deployment across various industries were subjected to critical examination based on the extant literature.\u003c/p\u003e \u003cp\u003e \u003cb\u003eModel Performance and Comparison with Previous Studies\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThis study introduces a novel approach to sentiment analysis by employing the BiLSTM model with BERT Transformer, which represents an advancement over the previous research. While earlier investigations \u0026lrm;[11\u0026lrm;,12] relied on logistic regression or machine learning algorithms, such as support vector machines and Naive Bayes, they failed to effectively capture long-term dependencies and contextual information in written data. The BiLSTM model with BERT (Hybrid method) utilized\u003c/p\u003e \u003cp\u003ein this research offers a sophisticated method for analyzing sentiment in text, considering both forward and backward aspects. Previous studies acknowledged the issue of bias in sentiment\u003c/p\u003e \u003cp\u003einterpretation but did not provide concrete solutions. To address this gap, our study implemented an effective bias mitigation strategy, including BERT Transformer, text normalization and stop\u003c/p\u003e \u003cp\u003eword removal, resulting in more precise outcomes. Additionally, while past research has primarily focused on improving accuracy, it has often overlooked other ethical considerations. Our approach advances the field by incorporating performance enhancement vertices and improvement techniques for bias mitigation, making it not only efficient, but also ethically sound for developing responsible AI systems.\u003c/p\u003e \u003cp\u003eOur research distinguishes itself from prior studies by emphasizing the significance of combining accuracy and context while using AI responsibly, thus addressing sentiment analysis from a more comprehensive perspective.\u003c/p\u003e \u003cp\u003e \u003cb\u003eBias and Ethical Considerations\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe ethical framework was created using the principles of fairness, accountability, and transparency (FAT). Its development involves three essential steps: (1) identifying ethical constraints unique to AI in each area, (2) examining existing guidelines such as the OECD AI principles, and (3) adapting recommendations to address sector-specific concerns. For example, in healthcare, the framework prioritizes patient data protection, whereas in education, it concentrates on eliminating plagiarism and ensuring equal access. These concepts were used to assess ChatGPT's compliance with ethical norms in the business, education, and healthcare sectors.\u003c/p\u003e \u003cp\u003eIn certain studies, \u0026lrm;[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e] shows the significance of addressing bias in artificial intelligence (AI) systems has been emphasized in the domains of healthcare and education, where inaccurate predictions can have substantial real-world implications because they are sensitive. With the incorporation of text processing into algorithms such as BILSTM with BERT, there has been a concerted effort to mitigate the bias from model predictions. Through the analysis and processing of texts to eliminate words that do not influence sentiment, this study aims to ensure that the model's predictions are not disproportionately affected by specific phrases or linguistic patterns. This approach is crucial for reducing the biases that might otherwise emerge from the linguistic or demographic characteristics of the dataset.\u003c/p\u003e \u003cp\u003eHowever, although these measures contribute to fairness, they do not eliminate the risk of bias entirely. Ongoing investigations should focus on developing additional techniques to detect and address bias, ensuring that artificial intelligence systems deliver fair results for all individuals who use them. Subsequently, based on these results, it is imperative to elucidate the negative aspects\u003c/p\u003e \u003cp\u003eof ChatGPT, necessitating an understanding of the ethical implications associated with its use and potential mitigation strategies. Privacy concerns arise when individuals share personal information with ChatGPT for advice, as the AI model may inadvertently disclose sensitive details from the user's past, resulting in a breach of privacy. Furthermore, organizations may utilize ChatGPT to engage with customers and obtain their information without explicit consent, constituting an\u003c/p\u003e \u003cp\u003eunethical misuse of data and a violation of privacy. Additionally, ChatGPT can be inappropriately employed to fabricate or disseminate information with the intent of causing harm to individuals or exploiting the technology for academic misconduct in scientific and literary fields, such as publishing plagiarized research papers, thereby compromising integrity and intellectual honesty.\u003c/p\u003e \u003cp\u003eThe above-mentioned ethical implications necessitate the implementation of strategies to address them when utilizing the ChatGPT. These strategies include establishing policies for AI\u003c/p\u003e \u003cp\u003edevelopers to provide transparent information regarding data usage, collection, and storage; implementing systems that facilitate the auditing and tracking of AI decisions; and ensuring accountability and transparency. Furthermore, it is essential to enhance training data to mitigate biases and ensure integrity as well as to create mechanisms for users to report potentially harmful or unethical AI behaviors. Of paramount importance is the incorporation of human oversight as a fundamental component in reviewing and intervening in AI decision-making processes to prevent ethical concerns that may affect efficiency and productivity in the workplace. \u0026lrm;[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/p\u003e \u003cp\u003e \u003cb\u003eField-Specific Implications\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe specific implications of this study are extensive and have exerted a significant influence on numerous industries, including business, healthcare, finance, and education, which have experienced substantial impacts through the application of natural language processing and artificial intelligence models encompassing the development of sentiment analysis utilizing artificial intelligence by integrating advanced deep learning into text-analysis-based fields using BILSTM with BERT and enhancing prediction accuracy in sentiment analysis. This is beneficial in marketing, customer service, and social media analysis, enabling organizations to gain a more comprehensive understanding of consumer opinions and sentiments.\u003c/p\u003e \u003cp\u003eIn Addition to the mitigation of bias in artificial intelligence systems through the implementation of text-processing techniques, such as converting text to lowercase letters and removing superfluous words, has a particular impact on the healthcare, education, and financial services sectors, where biased predictions can have serious real-world consequences. Ensuring fairness in artificial intelligence models is crucial for providing equitable services across diverse demographics. However, the study acknowledges that these measures do not completely eliminate bias, indicating that continued research is necessary to further develop unbiased artificial intelligence systems.\u003c/p\u003e \u003cp\u003eLastly, ethical considerations in deploying AI by designating human resources as the primary decision-maker are of paramount importance. This is particularly critical in domains such as human resources, legal services, and public administration, where AI recommendations or decisions must be reviewed to prevent potential ethical or legal violations. Industries must establish robust oversight mechanisms to ensure that AI systems augment human decision-making, rather than supplant it in sensitive contexts.\u003c/p\u003e"},{"header":"5 Conclusion and Future Work recommendation","content":"\u003cp\u003eThis study aimed to analyze and predict individuals' sentiments towards ChatGPT without bias, examine its impact on various industries, and evaluate the ethical considerations that should be considered when utilizing ChatGPT, employing machine and deep learning algorithms in natural language processing. The results of this investigation demonstrated that the Bidirectional LSTM with BERT model achieved significant superiority in performance and bias mitigation compared to machine learning algorithms with BERT owing to its ability to analyze data in multiple directions, yielding more accurate results. The study also explored how ethical considerations can be incorporated when using ChatGPT and its primary impacts on several sectors, including education, healthcare, and business. Consequently, future research should focus on developing additional techniques to further reduce bias in sentiment analysis, while emphasizing the importance of fairness and accuracy in sentiment prediction. Moreover, further research should address the broader ethical implications of deploying these models in sensitive industries such as healthcare and finance, where AI-driven decisions can have far-reaching consequences.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eEthical Consideration\u003c/h2\u003e \u003cp\u003eWe declare that all experiments and procedures conducted in this research adhere to the ethical standards outlined by Princess Sumaya University for Technology. Informed consent was obtained from all participants involved in the study, and their anonymity and confidentiality were strictly maintained.\u003c/p\u003e \u003ch2\u003eConflicts of Interest:\u003c/h2\u003e \u003cp\u003eThe authors declare that they have no conflicts of interest that could influence the outcome or interpretation of the research presented in this paper.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding:\u003c/h2\u003e \u003cp\u003eNo funding was received to assist with the preparation of this manuscript.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eStatement of Author Contributions:The research methodology and conceptualization were spearheaded by Laila Malas, who also played a key role in drafting the introduction. Ahmad Shawaqfeh took charge of the literature review, offering crucial insights and scholarly context. The final review and refinement of the manuscript were handled by Ahmad Abushakra, who ensured its overall coherence, precision, and compliance with academic standards. The writing, editing, and final approval of the manuscript before submission involved the collaborative efforts of all authors.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThe Corresponding author hereby to acknowledge their profound appreciation to all co-authors for their substantial contributions to this manuscript. The collective expertise and collaborative efforts of these individuals were instrumental in the completion of this research. Mr. Ahmad Shawaqfeh whose manage the literature review and provide valuable insights that enrich in depth of the discussion and Dr.Ahmad Abushakra who review and edit the manuscript enhancing the coherence of the final draft from the initial stages of idea development to the final edits, their steadfast dedication has been crucial in completing this project. We are grateful for the insightful feedback and conversations that helped us navigate obstacles and enhance our concepts. This publication exemplifies the effectiveness of collaboration and collective academic endeavor. The invaluable input from our colleagues was essential to our success in this endeavor.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eData is provided within the manuscript or supplementary information files\u0026rdquo; offer from Kaggel website under the following link : https://www.kaggle.com/datasets/tariqsays/chatgpt-twitter-dataset\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSch\u0026ouml;nberger, M. (2023). ChatGPT in Higher Education: The Good, The Bad, and The University. Editorial Universitat Polit\u0026egrave;cnica de Val\u0026egrave;ncia. 331-338. https://doi.org/10.4995/HEAd23.2023.16174\u003c/li\u003e\n\u003cli\u003eKuzub, A. (2023, May 16). Should you be using ChatGPT? Experts say \u0026lsquo;yes,\u0026rsquo; but don\u0026rsquo;t confuse it with a friend. Northeastern Global News. https://news.northeastern.edu/2023/05/16/using-chat-gpt/\u003c/li\u003e\n\u003cli\u003eVettaFi, LLC. (n.d.). VettaFi. VettaFi. https://insights.roboglobal.com/chatgpt-fact-from-fiction-expert-insights-%20vs.-public-opinion.\u003c/li\u003e\n\u003cli\u003eZhan, X., Xu, Y., \u0026amp; Sarkadi, S. (2023, July 19). Deceptive AI ecosystems: The case of ChatGPT. Proceedings of the 5th International Conference on Conversational User Interfaces, 35, 1\u0026ndash;6. Presented at the CUI \u0026rsquo;23: ACM conference on Conversational User Interfaces, Eindhoven Netherlands. https://doi.org/10.1145/3571884.3603754\u003c/li\u003e\n\u003cli\u003eRivas, P., \u0026amp; Zhao, L. (2023). Marketing with ChatGPT: Navigating the ethical terrain of GPT-based chatbot technology. AI, 4(2), 375-384. https://doi.org/10.3390/ai4020019\u003c/li\u003e\n\u003cli\u003eAlieksieiev, M., \u0026amp; Kurenkov, V. (2024). The present and the future of artificial intelligence. In Collection of Scientific Papers \u0026laquo;\u0026Lambda;ΌГO\u0026Sigma;\u0026raquo; (pp. 231\u0026ndash;232). https://doi.org/10.36074/logos-24.05.2024.050\u003c/li\u003e\n\u003cli\u003eGeetha, V., Gomathy, C. K., Teja, R. S., \u0026amp; Aniketh, V. (2023). THE FUTURE OF AI: TRANSFORMING TOMORROW. INTERANTIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT, 07(11), 1\u0026ndash;11. https://doi.org/10.55041/ijsrem27039\u003c/li\u003e\n\u003cli\u003eChouhan, P., Paliwal, M., \u0026amp; Gupta, T. (2023). Integration of natural language processing with artificial intelligence: A review paper. Journal of Analysis and Computation, 17(2), 294\u0026ndash;300. https://doi.org/ 10.30696/jac.xvii.2.2023.294-300\u003c/li\u003e\n\u003cli\u003eLi, K.-C., Bu, Z.-J., Shahjalal, M., He, B.-X., Zhuang, Z.-F., Li, C., \u0026hellip; Liu, Z.-L. (2024). Performance of ChatGPT on Chinese master\u0026rsquo;s degree entrance examination in Clinical Medicine. PloS One, 19(4), e0301702. doi: 10.1371/journal.pone.0301702\u003c/li\u003e\n\u003cli\u003eWu, Y., Si, S., Zhang, Y., Gu, J., \u0026amp; Wosik, J. (2024). Evaluating the performance of chatgpt for spam email detection. arXiv preprint arXiv:2402.15537.\u003c/li\u003e\n\u003cli\u003eSharma, S., Aggarwal, R., \u0026amp; Kumar, M. (2023, April). Mining Twitter for Insights into ChatGPT Sentiment: A Machine Learning Approach. In 2023 International Conference on Distributed Computing and Electrical Circuits and Electronics (ICDCECE) (pp. 1-6). IEEE.\u003c/li\u003e\n\u003cli\u003eEsh, M. (2024). Sentiment Analysis in ChatGpt Interactions: Unraveling Emotional Dynamics, Model Evaluation, and User Engagement Insights. Technical Services Quarterly, 41(2), 160\u0026ndash;174. https://doi.org/10.1080/07317131.2024.2319972\u003c/li\u003e\n\u003cli\u003eTariq \u0026middot;, M. (2AD), \u0026ldquo;ChatGPT Twitter Dataset\u0026rdquo;, Kaggle, Data set, Kaggle.\u003c/li\u003e\n\u003cli\u003eOECD. (2024, May 3). OECD updates AI Principles to stay abreast of rapid technological developments. OECD. Retrieved from https://www.oecd.org/en/about/news/press-releases/2024/05/oecd-updates-ai-principles-to-stay-abreast-of-rapid-technological-developments.html.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"discover-artificial-intelligence","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diai","sideBox":"Learn more about [Discover Artificial Intelligence](https://www.springer.com/44163)","snPcode":"","submissionUrl":"","title":"Discover Artificial Intelligence","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Artificial Intelligence, ChatGPT, Sentiment Analysis, Prediction, Sustainability","lastPublishedDoi":"10.21203/rs.3.rs-5180421/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5180421/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eTechnological advancements have significantly progressed from the inception of computers and the internet to the rise of artificial intelligence (AI), profoundly impacting various sectors such as business, education, and healthcare. Among these advancements, ChatGPT has emerged as a prominent conversational AI tool, facilitating tasks such as sentiment prediction and sector-specific decision-making. This study aims to evaluate ChatGPT's performance in analyzing public sentiment using data from X (formerly Twitter) and to propose an ethical framework for its sector-specific applications. By combining the BERT transformer model with BiLSTM, Random Forest, and K-Nearest Neighbor algorithms, the study achieves a hybrid approach to sentiment prediction, demonstrating superior performance with an accuracy of 86.7%. Furthermore, the study explores ethical considerations, such as fairness, accountability, and transparency, to address ChatGPT's adoption across industries. The findings offer insights into both the technical and ethical dimensions of ChatGPT\u0026rsquo;s integration into business, education, and healthcare sectors, emphasizing its potential for sustainable and responsible use.\u003c/p\u003e","manuscriptTitle":"The Future of Artificial Intelligence: Evaluating ChatGPT's Performance in X Sentiment Prediction.","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-13 05:49:45","doi":"10.21203/rs.3.rs-5180421/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Accepted","date":"2024-12-16T05:10:03+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-12-11T14:15:51+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-12-05T03:47:45+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Artificial Intelligence","date":"2024-11-15T11:14:02+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-artificial-intelligence","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diai","sideBox":"Learn more about [Discover Artificial Intelligence](https://www.springer.com/44163)","snPcode":"","submissionUrl":"","title":"Discover Artificial Intelligence","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"7e7b36da-1e3b-45af-9f56-525a77677eef","owner":[],"postedDate":"December 13th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2024-12-23T16:10:47+00:00","versionOfRecord":{"articleIdentity":"rs-5180421","link":"https://doi.org/10.1007/s44163-024-00218-0","journal":{"identity":"discover-artificial-intelligence","isVorOnly":false,"title":"Discover Artificial Intelligence"},"publishedOn":"2024-12-18 15:57:42","publishedOnDateReadable":"December 18th, 2024"},"versionCreatedAt":"2024-12-13 05:49:45","video":"","vorDoi":"10.1007/s44163-024-00218-0","vorDoiUrl":"https://doi.org/10.1007/s44163-024-00218-0","workflowStages":[]},"version":"v1","identity":"rs-5180421","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5180421","identity":"rs-5180421","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}