A Collaborative and Real-Time Model for Trusties Content in Social Media

A Collaborative and Real-Time Model for Trusties Content in Social Media | 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 A Collaborative and Real-Time Model for Trusties Content in Social Media Lyazid HAMIMED, Mourad AMAD, Abdelmalek BOUDRIES This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4359937/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Recently, social media is becoming a stronger tool for spreading news in the world. These platforms make it easy for anyone to disseminate their ideas, flood the world by different types of information. In order to minimize the inconvenience of fake news inundation, most of the developed techniques aim at detecting fake news by exploring how they propagate on the social media. Minimizing the negative effect of this kind of information, needs stronger mechanisms to detect fake news at an early stage by focusing on their contents. This paper proposes a new model for trusties’ content in social media. Its basic idea consists of combining news content and their propagation behavior over the social network. This model simulation shows that the susceptible fake news can be accused at an early stage. The performance evaluations show that the results are globally satisfactory. Social networking Social media Fake news Negative effect social profiling news features Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction The recent dissemination explosion of information, particularly on social media, has strong influence on modern society. Be it negative or positive, this influence triggered the involvement of researchers to meet the different users’ needs of social platforms. The majority of the works in this direction aims at maximizing the positive effect of social media. This paper proposes a new collaborative and real time model, able to judge the validity of the information disseminated on social platforms. First, this assessment consists of defining a probability value based on the characteristics of the information in question, as well as those of its source. Once the information starts disseminating, this probability is adjusted according to the users’ reactions. The proposed model uses a changelog to keep track of the validity of the information over time. This logging greatly facilitates the mechanisms integration to execute the appropriate actions on the context of the information disseminated. The rest of this paper is organized as follows. Section 2 presents background and related work. Section 3 describes the proposed model. Section 4 presents evaluation and performances of the proposed model on a random social network. Section 5 draws up conclusions and future works. 2. Background and Related Work Automatic detection of false information is a long-standing and largely unsolved problem. Worse, recent developments in language modeling allow automatic generation of information. An approach that has recently attracted attention detects false information by using provenance based on stylometry; i.e., by tracing the style of information expression to its source and determining if the source is malicious. This has proven to be very effective under the assumption that legitimate information is produced by humans and false information is produced by automatic tools. One approach to automating the detection of false information is fact-checking. These detectors are motivated by the way humans validate information [ 1 ] as well as the falsification mechanisms. However, despite the importance of work in this area, existing detectors are not yet precise enough to fully automate the detection task [ 2 , 3 ]. Where the authors examined automated research for fact-checking stemming from natural language processing and related disciplines, unifying task formulations and methodologies across papers and authors. They also emphasized the use of evidence as an important distinguishing factor, touching on task formulations and methods. They proposed avenues for future NLP research on automated fact-checking. James Thorne et al [ 3 ], in their article, they introduced a new publicly accessible dataset for textual source verification called FEVER. This dataset comprises 185,445 claims generated by sentences extracted from Wikipedia, with subsequent verification while ignoring their sources. The claims are categorized as supported, refuted, or NotEnoughInfo. To characterize the dataset's challenge, they developed a pipeline approach and compared it to properly designed oracles. The highest precision they achieved in labeling a claim with the correct evidence is 31.87%, while if they disregard the evidence, they achieve 50.91%. They believe that FEVER represents a challenging benchmark that will contribute to advancing the verification of claims against textual sources. Recently, Zellers et al. [ 4 ] proposed an alternative approach, which is based on identifying the source. This approach assumes that falsity is determined by the source that generated the information. For example, one can assume that press articles from a reputable newspaper are more accurate than articles from a propaganda website [ 5 ]. Alternatively, the source can indicate whether the article was written by a human or generated automatically [ 6 , 7 , 8 ]. Zellers et al. [ 4 ] used a language model to extract the characteristics of the article, which can be traced back to the information source. These characteristics include n-gram frequencies, sentence structures and text consistency, among others [ 9 ]. Fact verification is closely linked to the task of detecting false information [ 10 , 11 ]. It involves recovering potential evidence for a claim and assessing the link between them [ 12 ]. A fact-checking system can be used for the detection of false news by validating each of the article’s allegations against a reliable source. However, the performance of current automatic models is still relatively weak [ 2 , 13 ]. Detecting and reducing the spread of fake news has thus become a crucial challenge for researchers [ 14 ]. In order to mitigate the negative effects of this phenomenon, here is an overview of the approaches and technics used in this area: Among the most well-known approaches, there is text analysis [ 15 ], where Natural Language Processing (NLP) techniques play a crucial role in detecting fake news. NLP algorithms [ 16 ] analyze the content of information disseminated on social networks and other textual data to identify linguistic patterns [ 17 ], inconsistencies, and reveal anomalies characteristic of fake news [ 18 ]. In the same approach, there is also sentiment analysis [ 19 ], where we assess the emotional tone of a part of the text. Fake news often uses emotionally charged language to manipulate readers, and this type of analysis can help detect such information. Named Entity Recognition (NER) [ 20 ] is also used in this first approach. NER involves identifying named entities such as individuals, places, and organizations. This can reveal patterns of deception when false sources misuse or create new entities. Source analysis is a highly effective approach in this context. In this approach, credibility is evaluated and considered fundamental for fake news detection [ 21 ]. The algorithms in this approach assess the reputation and history of media outlets, websites, and social media accounts to gauge their reliability. Network analysis [ 22 ], which involves examining social media connections and user interactions, can provide insights into suspicious leads and patterns, such as the spread of fake news from certain providers. Fact-checking and Verification This is an approach that involves collaborating with fact-checking organizations, where such collaboration can provide valuable information for verification purposes [ 23 ]. The algorithms in this approach cross-reference news articles with fact-checking databases to identify discrepancies. There is also multimodal verification, where fact-checking can extend beyond text by analyzing images, videos, and audio content to detect any manipulation or false statements. Machine Learning and Deep Learning The majority of works of this approach are based on connectionist artificial intelligence, where learning models are trained on labeled datasets of fake and real news to classify new articles [ 24 ]. Their uses encompass linguistic models, source credibility, and user interactions. Deep Learning is also employed in this approach, where deep neural networks, such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), have been applied to detect fake news by learning complex representations of textual and contextual information. Multiplatform Detection and Real-Time Detection In terms of Multiplatform Analysis, fake news typically spreads across multiple platforms [ 25 ]. Consequently, generalized detection systems are developed to analyze the dissemination of information across social networks, news websites, and forums. Real-time detection is a crucial factor in combating the rapid spread of fake news. Real-time detection systems continuously monitor and assess emerging news as it unfolds. Explainability and Bias Mitigation Algorithms of this approach provide and ensure transparency in fake news detection, which is crucial [ 26 ]. They offer explanations for their decisions to help users to understand why certain information is classified as false. Detecting and mitigating biases in fake news detection algorithms are essential to avoid false positives or negatives based on political, cultural, or linguistic prejudices. Overall, fake news detection approaches are continually evolving, combining various NLP, machine learning, and source analysis techniques [ 27 ] to address the complex challenges posed by misinformation in today's information landscape. Researchers and practitioners are diligently working to develop more robust and accurate methods for identifying and combating fake news. 3. Proposed model This section details the proposed model for supervising news trusties over the social media. a. Global model view Figure 1 outlines main entities of a social media, where users can inject news enriching the media social’s capital and performing the social computing that can requiring external context’s information. b. Basic idea Since its creation by its owner at time \({t}_{0}\) , the news \({N}_{i}\) is propagated from a level to another through social links (Level 0 if the information is adopted only by its owner; the next levels for those who adopted it and are friends of the owner). At level \(k\) , the information is adopted by a subset of users labeled \({\cup }_{{N}_{i}}^{k}=\left\{{U}_{1}, ..., {U}_{m}\right\}\) . This propagation can be defined by a propagation function: from \({\cup }_{{N}_{i}}^{k}\) , the set of users at level \({L}_{k}\) , to \({\cup }_{{N}_{i}}^{k+1}\) , the next level \({\text{L}}_{k+1,k\ge 1}\) of users \(\) . At level \({L}_{0}\) , only the owner of the news adopts the information in question, \({\cup }_{Ni}^{0}=\left\{Owner\right({N}_{i}\left)\right\}\) . The recursive definition of sets \({ \cup }_{Ni}^{n}\) ; i.e., users adopting the information in question at the next level are given by formula (1): \({\cup }_{{N}_{i}}^{k+1}={\cup }_{Ni}^{k}{+ U}_{i} of the Network \nexists {U}_{j}\in {\cup }_{Ni}^{k}\) \(where{ U}_{j} is in relation whith{ U}_{i}\) (1) c. Evaluation of the validity of news For each level of information propagation, the validity of the information is estimated as a probability. This probability is initialized at level \({L}_{0}\) then adjusted according to users’ reactions and features to the next level. At level \({L}_{i}\) it can be calculated by formula 2. \({P}_{N}^{i}=\frac{\sum {\text{U}}_{\text{k}}\in {\cup }_{N}^{i}, \text{P}({\text{U}}_{\text{K}},\text{N}\text{e}\text{w}\text{s})}{\left|{\cup }_{N}^{i}\text{i}\right|}\) (2) where \(P\left({U}_{K},News\right)\) is computed based on the profile user’s \({U}_{k}\) and theirs reactions as well as the characteristics of News \(N\) . \({U}_{k}\) represents all \({\cup }_{N}^{i}\) users adopting the news \(N\) at level \({L}_{i}\) . d. Model formulation and representation of social entities Before proposing necessary algorithms for implementing the proposed model, it is necessary to describe the active entities in the social media (such as the representation of information disseminated in these media, its actors as well as their relationships, etc.). e. News The information disseminated in the online social networks is identified by a unique identifier ID, as it is characterized by a subset of possible features. It is represented by a pair composed of identification and a vector of its characteristics. With the scale of Dataset, hash functions can be used to facilitate indexing process. \(News=\left\{\begin{array}{c}News\_ID \\ Features=\left[{f}_{1},{f}_{2},\dots ,{f}_{n}\right], {f}_{i}\in \left[0, 1\right]\end{array}\right.\) (3) f. Users Social media users are identified by their unique IDs and are specialized in a subset of different domains with a certain percentage. Each user of the social network is represented by a couple composed by his identification and all specialization areas. \({U}_{S}=\left\{\begin{array}{c} U{s}_{ID}\\ Spe=\left[{S}_{1},{S}_{2},\dots ,{S}_{n}\right], {S}_{i}\in \left[0, 1\right]\end{array}\right.\) (4) 4. Performance and evaluation To simulate the proposed model and evaluate its efficiency, an environment is developed under Python language. This implementation is based on the networkx library . This section presents algorithms and principal simulation results. a. Algorithms and python code The social network is represented by a graph \(G(U,V)\) where \(U\) is a set of network users and \(V\) represents users relationships of the network. This graph is generated randomly, which is the result of the following Python code: G = nx.gnm_random_graph(nb_nodes, nb_edges) Next, users and posts features and users adopting each post at different levels are generated by the following Python code: add_features_nodes("features_file_name") data_generation(10, deep_netx) graph_adopts_generation(deep_netx) Then, posts trusties are calculated as follows: posts_prob_level_i("proba_all_levels.txt", deep_netx) Post trustiest G(U,V) the social network; Posts: posts diffused over G; For post in Posts: #according to post features and user profile Initialize_trustes(post) Update_Trustees_Post (post, G_post, level_i); #where G_post is the subgraph of G infected by post b. Network initialization To evaluate the efficiency of the proposed model, a social network is generated. Here, a set of features is assigned to each user, which describes user specialization. Value of each feature is in range [0, 1], which can easily represents probability notion. Figure 2 shows a graph \(G(10, 30)\) that simulates a social network of 10 users and 30 relationships. This network is generated randomly by the developed simulator, where users are numbered from 0 to 9 as well as for each user, their features as a percentage in the interval [0, 1]. Figure 3 represents another example of a social network to evaluate the proposed model using the power of user links. In this graph, consisting of 5 users and 8 relationships, the connections between users are labeled with a value ranging from 0 to 1. This value represents the strength of the link between the two users involved in such a relationship. For example, the relationship between the user at node 0 and the one at node 4 is very weak, whereas a very significant relationship can be observed between the same user and the one at node 3. The histogram of Fig. 4 shows the influence of the user’s profile on the validity factor of the diffused post at level 0 (when the same post is broadcasted by different users of the network). These results illustrate how the validity indicator varies depending on the characteristics of the disseminated information and the expertise of the author. At the moment of dissemination, these findings reveal that User 2 specializes in the subject matter of the information being shared, where the system assigns a validation probability of 0.51. In contrast, if User 3 disseminates the information, the system assigns a validation probability of 0.16. Table 1 contains a social capital of 5 posts diffused by users of the generated network and the features of same post. Table 1 Post’s ID, features and trusties at level 0 Post_ID User_ID F0 F1 F2 F3 Trust_level_0 0 2 0.7 0.9 0.2 1 0.51 1 0 0.7 0.9 0.2 1 0.45 2 3 0.7 0.9 0.2 1 0.16 3 1 0.7 0.9 0.2 1 0.188 4 4 0.7 0.9 0.2 1 0.335 This table shows the probability of validation, at the time of broadcasting, for 5 pieces of information with the same characteristics on the social network. For example, information 1, with the characteristics (0.7, 0.9, 0.2, 1), if broadcasted by user 0, will have a validity probability of 0.45. c. Trusties supervision of posts over the network Trusties’ supervision under the developed environment consists of displaying how indicator post’s evolves over the time for each post. Example G(20,50) : To generate this graph (Fig. 5 ), this python code is executed: G = nx.gnm_random_graph(nb_nodes = 20, nb_edges = 50) Over the network of Fig. 5 , Table 2 shows the evolution of the probability of validity of each post, according to the level of its propagation in the social network. Table 2 Posts trusties at each level deep0 deep1 deep2 deep3 post1 0.202 0.2083125 0.2175 0.237928571 post2 0.541 0.3803125 0.316944444 0.298871622 post3 0.40225 0.379166667 0.3476625 0.337692308 post4 0.466 0.416875 0.387096154 0.381377778 post5 0.209 0.206975 0.207607143 0.209595745 post6 0.10975 0.190958333 0.233590909 0.239518293 post7 0.115 0.09705 0.109592105 0.117756579 post8 0.12625 0.12375 0.112208333 0.104952703 post9 0.09675 0.25734375 0.279407407 0.283244565 post10 0.1035 0.233708333 0.268180556 0.287695946 The graph of Fig. 5 clearly shows the evolution of this probability according to the spread of the posts in the social media. For example, post2 has a behavior of suspicious fake news broadcasted by a user who specializes in this news field, where its truth indicator quickly decreases, where its truth indicator quickly decreases. On the other hand, post10 behaves like a real post, broadcasted by a user who is not specialized in the content of this post. However, its probability (0.1035) at deep0 was visibly low, then it more than doubled at the next level (deep1), and continued to increase for the subsequent levels (deep2 and deep3). This can be justified by positive reactions of users at different levels (deep1, deep2, and deep3). The graph of Fig. 7 is another representation of the evolution of the post’s trusty’s indicator. In this graph, it can be noticed that post1 , post5 , post7 and post8 have certain stability over the social network. This allows to deduce that these posts can be trusted. Indicators of trusty for certain posts such as post1, post6, post9, and post10, continue to increase as they spread further through the social network. This increase signifies positive backing from participants within the social network for these posts, consequently classifying them as genuine. On the other hand, trusties’ factor for posts post2, post3, post4, and post8 continues to decrease as they propagate through the network. This indicates the adverse responses from social media users during the dissemination of these posts through the social network. 5. Conclusion and future works This work presented a new model for fake news detection in social media. For this purpose, it presented the overall behavior of a social media that can decipher fake news early as possible. This model represents the social capital by adequate concepts allowing adaptation in new technologies. The prediction mechanism is based on a probabilistic model, where users and their posts, over the social media, are modeled in this concept. The implicit supervision of the validity of a post in real-time and simulation results are promising. Future work will deal the implementation of this work over a real social network based on the IPFS distributed system. Declarations The authors have no relevant financial or non-financial interests to disclose. Author Contribution Lyazid HAMIMED wrote the main manuscript text, and all authors reviewed the manuscript.This work is a contribution of the PhD student Lyazid HAMIMED, supervised by Professor Mourad AMAD and co-supervised by Abdelmalek BOUDRIES. References Nyhan Brendan, Reifler Jason. Estimating fact-checking’s effects , Arlington, VA: American Press Institute. 2015. Thorne James, Vlachos Andreas. Automated Fact Checking: Task Formulations, Methods and Future Directions , Proceedings of the 27th International Conference on Computational Linguistics. 2018. 3346–3359. Thorne James, Vlachos Andreas, Christodoulopoulos Christos, Mittal Arpit. FEVER: a Large-scale Dataset for Fact Extraction and VERification , Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018. 809–819. Zellers Rowan, Holtzman Ari, Rashkin Hannah, Bisk Yonatan, Farhadi Ali, Roesner Franziska, Choi Yejin. Defending Against Neural Fake News , arXiv preprint arXiv:1905.12616. 2019. Baly Ramy, Karadzhov Georgi, Alexandrov Dimitar, Glass James, Nakov Preslav. Predicting Factuality of Reporting and Bias of News Media Sources , Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Brussels, Belgium: Association for Computational Linguistics, X-XI 2018. 3528–3539. Hashimoto Tatsunori, Zhang Hugh, Liang Percy. Unifying Human and Statistical Evaluation for Natural Language Generation , Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). Minneapolis, Minnesota: Association for Computational Linguistics, VI 2019. 1689–1701. Gehrmann Sebastian, Strobelt Hendrik, Rush Alexander. GLTR: Statistical Detection and Visualization of Generated Text , Proceedings of the 57th Conference of the Association for Computational Linguistics: System Demonstrations. Florence, Italy: Association for Computational Linguistics, VII 2019. 111–116. Bakhtin Anton, Gross Sam, Ott Myle, Deng Yuntian, Ranzato Marc’Aurelio, Szlam Arthur. Real or Fake? Learning to Discriminate Machine from Human Generated Text , arXiv preprint arXiv:1906.03351. 2019. Pérez-Rosas Verónica, Kleinberg Bennett, Lefevre Alexandra, Mihalcea Rada. Automatic Detection of Fake News , Proceedings of the 27th International Conference on Computational Linguistics. 2018. 3391–3401. Vlachos Andreas, Riedel Sebastian. Fact Checking: Task definition and dataset construction , Proceedings of the ACL 2014 Workshop on Language Technologies and Computational Social Science. Baltimore, MD, USA: Association for Computational Linguistics, VI 2014. 18–22. Wang William Yang. “ Liar, Liar Pants on Fire”: A New Benchmark Dataset for Fake News Detection , Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers). Vancouver, Canada: Association for Computational Linguistics, VII 2017. 422–426. Popat Kashyap, Mukherjee Subhabrata, Strötgen Jannik, Weikum Gerhard. Where the Truth Lies: Explaining the Credibility of Emerging Claims on the Web and Social Media, Proceedings of the 26th International Conference on World Wide Web Companion. Republic and Canton of Geneva, Switzerland: International World Wide Web Conferences Steering Committee, 2017. 1003–1012. (WWW ‘17 Companion). Schuster Tal, Shah Darsh J, Yeo Yun Jie Serene, Filizzola Daniel, Santus Enrico, Barzilay Regina. Towards Debiasing Fact Verification Models , Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. 2019. Á. Figueira et L. Oliveira, « The current state of fake news: challenges and opportunities », Procedia Computer Science, vol. 121, p. 817‑825, janv. 2017, doi: 10.1016/j.procs.2017.11.106. H. Himdi, G. Weir, F. Assiri, et H. Al-Barhamtoshy, « Arabic Fake News Detection Based on Textual Analysis », Arab J Sci Eng, vol. 47, no 8, p. 10453‑10469, août 2022, doi: 10.1007/s13369-021-06449-y. C. Busioc, S. Ruseti, et M. Dascalu, « A Literature Review of NLP Approaches to Fake News Detection and Their Applicability to Romanian-Language News Analysis », Transilvania, p. 65‑71, oct. 2020, doi: 10.51391/trva.2020.10.07. S. C. Guntuku, D. B. Yaden, M. L. Kern, L. H. Ungar, et J. C. Eichstaedt, « Detecting depression and mental illness on social media: an integrative review », Current Opinion in Behavioral Sciences, vol. 18, p. 43‑49, déc. 2017, doi: 10.1016/j.cobeha.2017.07.005. M. Aldwairi et A. Alwahedi, « Detecting Fake News in Social Media Networks », Procedia Computer Science, vol. 141, p. 215‑222, janv. 2018, doi: 10.1016/j.procs.2018.10.171. M. A. Alonso, D. Vilares, C. Gómez-Rodríguez, et J. Vilares, « Sentiment Analysis for Fake News Detection », Electronics, vol. 10, no 11, Art. no 11, janv. 2021, doi: 10.3390/electronics10111348. G. De Magistris, S. Russo, P. Roma, J. T. Starczewski, et C. Napoli, « An Explainable Fake News Detector Based on Named Entity Recognition and Stance Classification Applied to COVID-19 », Information, vol. 13, no 3, Art. no 3, mars 2022, doi: 10.3390/info13030137. X. Zhang et A. A. Ghorbani, « An overview of online fake news: Characterization, detection, and discussion », Information Processing & Management, vol. 57, no 2, p. 102025, mars 2020, doi: 10.1016/j.ipm.2019.03.004. K. Shu, H. R. Bernard, et H. Liu, « Studying Fake News via Network Analysis: Detection and Mitigation », in Emerging Research Challenges and Opportunities in Computational Social Network Analysis and Mining, N. Agarwal, N. Dokoohaki, et S. Tokdemir, Éd., in Lecture Notes in Social Networks. , Cham: Springer International Publishing, 2019, p. 43‑65. doi: 10.1007/978-3-319-94105-9_3. A. Ardèvol-Abreu, P. Delponti, et C. Rodríguez-Wangüemert, « Intentional or inadvertent fake news sharing? Fact-checking warnings and users’ interaction with social media content », Profesional de la información, vol. 29, no 5, Art. no 5, sept. 2020, doi: 10.3145/epi.2020.sep.07. « Fake News Detection Using Machine Learning and Deep Learning Algorithms ». Consulté le: 27 septembre 2023. [En ligne]. Disponible sur: https://ieeexplore.ieee.org/abstract/document/9436605/ Z. Chen et al., « An Automatic Framework to Continuously Monitor Multi-Platform Information Spread », août 2021. K. Shu, L. Cui, S. Wang, D. Lee, et H. Liu, « dEFEND: Explainable Fake News Detection », in Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, in KDD ’19. New York, NY, USA: Association for Computing Machinery, juill. 2019, p. 395‑405. doi: 10.1145/3292500.3330935. S. Ahmed, K. Hinkelmann, et F. Corradini, « Development of Fake News Model using Machine Learning through Natural Language Processing ». arXiv, 19 janvier 2022. doi: 10.48550/arXiv.2201.07489. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4359937","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":301566266,"identity":"644cb6fd-a953-48a6-8888-a4618b60099b","order_by":0,"name":"Lyazid HAMIMED","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6UlEQVRIiWNgGAWjYBACAwjJLMfGzNj4gIHhAJDH+IAoLcb87M2HDSBamA2I0MLAnDiz51iaBFFazBl4DD/+KLBm3HAjx6yap+aOHD8DM9sHfFosG3iMJSQM0pkNgFpu8xx7ZizZwMw8A6/DDvAYSBgYHGaDaGE7nLjhAP9h/H45wGP8I8HgMA9ISzHPP5AWZmZCWswkDhgclpAEep+Zt40oLWxllg0G6QagQJac23fYWLKZoBbmzTd//LGubwNG5Yc33w7LAfXi18Ig/wIRC0w8IJKABiBgfwBnMv4gqHoUjIJRMApGIgAAdUNIzuua5qUAAAAASUVORK5CYII=","orcid":"","institution":"University of Béjaïa","correspondingAuthor":true,"prefix":"","firstName":"Lyazid","middleName":"","lastName":"HAMIMED","suffix":""},{"id":301566267,"identity":"5c9e2834-1b48-4c0c-babe-b8627161e982","order_by":1,"name":"Mourad AMAD","email":"","orcid":"","institution":"University of Bouira","correspondingAuthor":false,"prefix":"","firstName":"Mourad","middleName":"","lastName":"AMAD","suffix":""},{"id":301566268,"identity":"f2132689-a48b-479b-993b-b8294f01d180","order_by":2,"name":"Abdelmalek BOUDRIES","email":"","orcid":"","institution":"University of Béjaïa","correspondingAuthor":false,"prefix":"","firstName":"Abdelmalek","middleName":"","lastName":"BOUDRIES","suffix":""}],"badges":[],"createdAt":"2024-05-02 15:15:00","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-4359937/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4359937/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":56430239,"identity":"6583556b-bafc-4fb5-8b67-250e0f2ff7e8","added_by":"auto","created_at":"2024-05-14 05:58:06","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":219221,"visible":true,"origin":"","legend":"\u003cp\u003eOverall diagram of the proposed model\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4359937/v1/3f9a124c22ecc4c16175b8f7.jpg"},{"id":56430241,"identity":"9fb8d1ea-66db-47f8-a5f9-cbc22384f586","added_by":"auto","created_at":"2024-05-14 05:58:07","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":89407,"visible":true,"origin":"","legend":"\u003cp\u003eA social network with user’s features\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-4359937/v1/2750e2315644de30201bedc9.png"},{"id":56430242,"identity":"59d58724-6622-4d1e-bcf0-489f863d1e1c","added_by":"auto","created_at":"2024-05-14 05:58:07","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":51266,"visible":true,"origin":"","legend":"\u003cp\u003eA social network with power links\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-4359937/v1/e76b241c258c9c4d0c64c4a3.png"},{"id":56430245,"identity":"ea1b7f02-3cd8-4a69-9625-541208acfc26","added_by":"auto","created_at":"2024-05-14 05:58:08","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":9789,"visible":true,"origin":"","legend":"\u003cp\u003eMeans validity factor at level 0\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-4359937/v1/33d1904680e034449ed9bd44.png"},{"id":56430243,"identity":"b5246e68-20af-46ac-9397-43c3d05ca194","added_by":"auto","created_at":"2024-05-14 05:58:07","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":186886,"visible":true,"origin":"","legend":"\u003cp\u003eExample of a social network G(20, 50) generated randomly\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-4359937/v1/8b34e8bac425438ba7a9e8b2.png"},{"id":56430246,"identity":"df1a24c6-cd36-41e6-8bb8-c0a14d848031","added_by":"auto","created_at":"2024-05-14 05:58:08","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":310157,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePost’s trusties evolution by level\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4359937/v1/618e336ab113d3b6a89902a2.jpg"},{"id":56430244,"identity":"f078cd7e-cdb0-4395-853b-11e14b63b76c","added_by":"auto","created_at":"2024-05-14 05:58:07","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":155091,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePosts and Trusties evolution\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4359937/v1/65d1af9893d64dff13283d28.jpg"},{"id":56431003,"identity":"f106993b-cb83-49e9-9414-641c98376d38","added_by":"auto","created_at":"2024-05-14 06:14:15","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1551433,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4359937/v1/0b544081-6539-4f7c-856e-881a27732bec.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Collaborative and Real-Time Model for Trusties Content in Social Media","fulltext":[{"header":"1.\tIntroduction","content":"\u003cp\u003eThe recent dissemination explosion of information, particularly on social media, has strong influence on modern society. Be it negative or positive, this influence triggered the involvement of researchers to meet the different users\u0026rsquo; needs of social platforms. The majority of the works in this direction aims at maximizing the positive effect of social media.\u003c/p\u003e\n\u003cp\u003eThis paper proposes a new collaborative and real time model, able to judge the validity of the information disseminated on social platforms. First, this assessment consists of defining a probability value based on the characteristics of the information in question, as well as those of its source. Once the information starts disseminating, this probability is adjusted according to the users\u0026rsquo; reactions. The proposed model uses a changelog to keep track of the validity of the information over time. This logging greatly facilitates the mechanisms integration to execute the appropriate actions on the context of the information disseminated.\u003c/p\u003e\n\u003cp\u003eThe rest of this paper is organized as follows. Section 2 presents background and related work. Section 3 describes the proposed model. Section 4 presents evaluation and performances of the proposed model on a random social network. Section 5 draws up conclusions and future works.\u003c/p\u003e"},{"header":"2. Background and Related Work","content":"\u003cp\u003eAutomatic detection of false information is a long-standing and largely unsolved problem. Worse, recent developments in language modeling allow automatic generation of information.\u003c/p\u003e \u003cp\u003eAn approach that has recently attracted attention detects false information by using provenance based on stylometry; i.e., by tracing the style of information expression to its source and determining if the source is malicious. This has proven to be very effective under the assumption that legitimate information is produced by humans and false information is produced by automatic tools. One approach to automating the detection of false information is fact-checking. These detectors are motivated by the way humans validate information [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e] as well as the falsification mechanisms. However, despite the importance of work in this area, existing detectors are not yet precise enough to fully automate the detection task [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Where the authors examined automated research for fact-checking stemming from natural language processing and related disciplines, unifying task formulations and methodologies across papers and authors. They also emphasized the use of evidence as an important distinguishing factor, touching on task formulations and methods. They proposed avenues for future NLP research on automated fact-checking. James Thorne et al [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], in their article, they introduced a new publicly accessible dataset for textual source verification called FEVER. This dataset comprises 185,445 claims generated by sentences extracted from Wikipedia, with subsequent verification while ignoring their sources. The claims are categorized as supported, refuted, or NotEnoughInfo. To characterize the dataset's challenge, they developed a pipeline approach and compared it to properly designed oracles. The highest precision they achieved in labeling a claim with the correct evidence is 31.87%, while if they disregard the evidence, they achieve 50.91%. They believe that FEVER represents a challenging benchmark that will contribute to advancing the verification of claims against textual sources.\u003c/p\u003e \u003cp\u003eRecently, Zellers et al. [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] proposed an alternative approach, which is based on identifying the source. This approach assumes that falsity is determined by the source that generated the information. For example, one can assume that press articles from a reputable newspaper are more accurate than articles from a propaganda website [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Alternatively, the source can indicate whether the article was written by a human or generated automatically [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eZellers et al. [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] used a language model to extract the characteristics of the article, which can be traced back to the information source. These characteristics include \u003cem\u003en-gram\u003c/em\u003e frequencies, sentence structures and text consistency, among others [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFact verification is closely linked to the task of detecting false information [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. It involves recovering potential evidence for a claim and assessing the link between them [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. A fact-checking system can be used for the detection of false news by validating each of the article\u0026rsquo;s allegations against a reliable source. However, the performance of current automatic models is still relatively weak [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDetecting and reducing the spread of fake news has thus become a crucial challenge for researchers [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. In order to mitigate the negative effects of this phenomenon, here is an overview of the approaches and technics used in this area:\u003c/p\u003e \u003cp\u003eAmong the most well-known approaches, there is \u003cb\u003etext analysis\u003c/b\u003e [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], where Natural Language Processing (NLP) techniques play a crucial role in detecting fake news. NLP algorithms [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] analyze the content of information disseminated on social networks and other textual data to identify linguistic patterns [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e], inconsistencies, and reveal anomalies characteristic of fake news [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. In the same approach, there is also sentiment analysis [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e], where we assess the emotional tone of a part of the text. Fake news often uses emotionally charged language to manipulate readers, and this type of analysis can help detect such information. Named Entity Recognition (NER) [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] is also used in this first approach. NER involves identifying named entities such as individuals, places, and organizations. This can reveal patterns of deception when false sources misuse or create new entities.\u003c/p\u003e \u003cp\u003e \u003cb\u003eSource analysis\u003c/b\u003e is a highly effective approach in this context. In this approach, credibility is evaluated and considered fundamental for fake news detection [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. The algorithms in this approach assess the reputation and history of media outlets, websites, and social media accounts to gauge their reliability. Network analysis [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], which involves examining social media connections and user interactions, can provide insights into suspicious leads and patterns, such as the spread of fake news from certain providers.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eFact-checking and Verification\u003c/strong\u003e \u003cp\u003eThis is an approach that involves collaborating with fact-checking organizations, where such collaboration can provide valuable information for verification purposes [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. The algorithms in this approach cross-reference news articles with fact-checking databases to identify discrepancies. There is also multimodal verification, where fact-checking can extend beyond text by analyzing images, videos, and audio content to detect any manipulation or false statements.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eMachine Learning and Deep Learning\u003c/strong\u003e \u003cp\u003eThe majority of works of this approach are based on connectionist artificial intelligence, where learning models are trained on labeled datasets of fake and real news to classify new articles [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Their uses encompass linguistic models, source credibility, and user interactions. Deep Learning is also employed in this approach, where deep neural networks, such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), have been applied to detect fake news by learning complex representations of textual and contextual information.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eMultiplatform Detection and Real-Time Detection\u003c/strong\u003e \u003cp\u003eIn terms of Multiplatform Analysis, fake news typically spreads across multiple platforms [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Consequently, generalized detection systems are developed to analyze the dissemination of information across social networks, news websites, and forums. Real-time detection is a crucial factor in combating the rapid spread of fake news. Real-time detection systems continuously monitor and assess emerging news as it unfolds.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eExplainability and Bias Mitigation\u003c/strong\u003e \u003cp\u003eAlgorithms of this approach provide and ensure transparency in fake news detection, which is crucial [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. They offer explanations for their decisions to help users to understand why certain information is classified as false. Detecting and mitigating biases in fake news detection algorithms are essential to avoid false positives or negatives based on political, cultural, or linguistic prejudices.\u003c/p\u003e \u003c/p\u003e \u003cp\u003eOverall, fake news detection approaches are continually evolving, combining various NLP, machine learning, and source analysis techniques [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] to address the complex challenges posed by misinformation in today's information landscape. Researchers and practitioners are diligently working to develop more robust and accurate methods for identifying and combating fake news.\u003c/p\u003e"},{"header":"3. Proposed model","content":"\u003cp\u003eThis section details the proposed model for supervising news trusties over the social media.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea. Global model view\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e outlines main entities of a social media, where users can inject news enriching the media social\u0026rsquo;s capital and performing the social computing that can requiring external context\u0026rsquo;s information.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eb. Basic idea\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSince its creation by its owner at time \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({t}_{0}\\)\u003c/span\u003e\u003c/span\u003e, the news \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({N}_{i}\\)\u003c/span\u003e\u003c/span\u003e is propagated from a level to another through social links (Level 0 if the information is adopted only by its owner; the next levels for those who adopted it and are friends of the owner). At level \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(k\\)\u003c/span\u003e\u003c/span\u003e, the information is adopted by a subset of users labeled \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\cup }_{{N}_{i}}^{k}=\\left\\{{U}_{1}, ..., {U}_{m}\\right\\}\\)\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e\u003cp\u003eThis propagation can be defined by a propagation function: from \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\cup }_{{N}_{i}}^{k}\\)\u003c/span\u003e\u003c/span\u003e, the set of users at level \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({L}_{k}\\)\u003c/span\u003e\u003c/span\u003e, to \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\cup }_{{N}_{i}}^{k+1}\\)\u003c/span\u003e\u003c/span\u003e, the next level \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\text{L}}_{k+1,k\\ge 1}\\)\u003c/span\u003e\u003c/span\u003eof users\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\)\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003eAt level \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({L}_{0}\\)\u003c/span\u003e\u003c/span\u003e, only the owner of the news adopts the information in question, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\cup }_{Ni}^{0}=\\left\\{Owner\\right({N}_{i}\\left)\\right\\}\\)\u003c/span\u003e\u003c/span\u003e. The recursive definition of sets\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({ \\cup }_{Ni}^{n}\\)\u003c/span\u003e\u003c/span\u003e; i.e., users adopting the information in question at the next level are given by formula (1):\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Taba\" border=\"1\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\cup }_{{N}_{i}}^{k+1}={\\cup }_{Ni}^{k}{+ U}_{i} of the Network \\nexists {U}_{j}\\in {\\cup }_{Ni}^{k}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\n \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(where{ U}_{j} is in relation whith{ U}_{i}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(1)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003ec. Evaluation of the validity of news\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor each level of information propagation, the validity of the information is estimated as a probability. This probability is initialized at level \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({L}_{0}\\)\u003c/span\u003e\u003c/span\u003e then adjusted according to users\u0026rsquo; reactions and features to the next level. At level \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({L}_{i}\\)\u003c/span\u003e\u003c/span\u003e it can be calculated by formula 2.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tabb\" border=\"1\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({P}_{N}^{i}=\\frac{\\sum {\\text{U}}_{\\text{k}}\\in {\\cup }_{N}^{i}, \\text{P}({\\text{U}}_{\\text{K}},\\text{N}\\text{e}\\text{w}\\text{s})}{\\left|{\\cup }_{N}^{i}\\text{i}\\right|}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(P\\left({U}_{K},News\\right)\\)\u003c/span\u003e\u003c/span\u003e is computed based on the profile user\u0026rsquo;s \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({U}_{k}\\)\u003c/span\u003e\u003c/span\u003e and theirs reactions as well as the characteristics of News \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(N\\)\u003c/span\u003e\u003c/span\u003e. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({U}_{k}\\)\u003c/span\u003e\u003c/span\u003e represents all \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\cup }_{N}^{i}\\)\u003c/span\u003e\u003c/span\u003e users adopting the news \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(N\\)\u003c/span\u003e\u003c/span\u003e at level \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({L}_{i}\\)\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ed. Model formulation and representation of social entities\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBefore proposing necessary algorithms for implementing the proposed model, it is necessary to describe the active entities in the social media (such as the representation of information disseminated in these media, its actors as well as their relationships, etc.).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ee. News\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe information disseminated in the online social networks is identified by a unique identifier ID, as it is characterized by a subset of possible features. It is represented by a pair composed of identification and a vector of its characteristics. With the scale of Dataset, hash functions can be used to facilitate indexing process.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tabc\" border=\"1\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(News=\\left\\{\\begin{array}{c}News\\_ID \\\\ Features=\\left[{f}_{1},{f}_{2},\\dots ,{f}_{n}\\right], {f}_{i}\\in \\left[0, 1\\right]\\end{array}\\right.\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(3)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003ef. Users\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSocial media users are identified by their unique IDs and are specialized in a subset of different domains with a certain percentage. Each user of the social network is represented by a couple composed by his identification and all specialization areas.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tabd\" border=\"1\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({U}_{S}=\\left\\{\\begin{array}{c} U{s}_{ID}\\\\ Spe=\\left[{S}_{1},{S}_{2},\\dots ,{S}_{n}\\right], {S}_{i}\\in \\left[0, 1\\right]\\end{array}\\right.\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e(4)\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e"},{"header":"4. Performance and evaluation","content":"\u003cp\u003eTo simulate the proposed model and evaluate its efficiency, an environment is developed under Python language. This implementation is based on the \u003cem\u003enetworkx library\u003c/em\u003e. This section presents algorithms and principal simulation results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea. Algorithms and python code\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe social network is represented by a graph \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(G(U,V)\\)\u003c/span\u003e\u003c/span\u003e where \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(U\\)\u003c/span\u003e\u003c/span\u003e is a set of network users and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(V\\)\u003c/span\u003e\u003c/span\u003e represents users relationships of the network. This graph is generated randomly, which is the result of the following Python code:\u003c/p\u003e\n\u003cp\u003eG\u0026thinsp;=\u0026thinsp;nx.gnm_random_graph(nb_nodes, nb_edges)\u003c/p\u003e\n\u003cp\u003eNext, users and posts features and users adopting each post at different levels are generated by the following Python code:\u003c/p\u003e\n\u003cp\u003eadd_features_nodes(\"features_file_name\")\u003c/p\u003e\n\u003cp\u003edata_generation(10, deep_netx)\u003c/p\u003e\n\u003cp\u003egraph_adopts_generation(deep_netx)\u003c/p\u003e\n\u003cp\u003eThen, posts trusties are calculated as follows:\u003c/p\u003e\n\u003cp\u003eposts_prob_level_i(\"proba_all_levels.txt\", deep_netx)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePost trustiest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eG(U,V) the social network;\u003c/p\u003e\n\u003cp\u003ePosts: posts diffused over G;\u003c/p\u003e\n\u003cp\u003eFor post in Posts:\u003c/p\u003e\n\u003cp\u003e#according to post features and user profile\u003c/p\u003e\n\u003cp\u003eInitialize_trustes(post)\u003c/p\u003e\n\u003cp\u003eUpdate_Trustees_Post (post, G_post, level_i);\u003c/p\u003e\n\u003cp\u003e#where G_post is the subgraph of G infected by post\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eb. Network initialization\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo evaluate the efficiency of the proposed model, a social network is generated. Here, a set of features is assigned to each user, which describes user specialization. Value of each feature is in range [0, 1], which can easily represents probability notion.\u003c/p\u003e\n\u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e shows a graph \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(G(10, 30)\\)\u003c/span\u003e\u003c/span\u003e that simulates a social network of 10 users and 30 relationships. This network is generated randomly by the developed simulator, where users are numbered from 0 to 9 as well as for each user, their features as a percentage in the interval [0, 1].\u003c/p\u003e\n\u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e represents another example of a social network to evaluate the proposed model using the power of user links. In this graph, consisting of 5 users and 8 relationships, the connections between users are labeled with a value ranging from 0 to 1. This value represents the strength of the link between the two users involved in such a relationship. For example, the relationship between the user at node 0 and the one at node 4 is very weak, whereas a very significant relationship can be observed between the same user and the one at node 3.\u003c/p\u003e\n\u003cp\u003eThe histogram of Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e shows the influence of the user\u0026rsquo;s profile on the validity factor of the diffused post at level 0 (when the same post is broadcasted by different users of the network).\u003c/p\u003e\n\u003cp\u003eThese results illustrate how the validity indicator varies depending on the characteristics of the disseminated information and the expertise of the author. At the moment of dissemination, these findings reveal that User 2 specializes in the subject matter of the information being shared, where the system assigns a validation probability of 0.51. In contrast, if User 3 disseminates the information, the system assigns a validation probability of 0.16.\u003c/p\u003e\n\u003cp\u003eTable 1 contains a social capital of 5 posts diffused by users of the generated network and the features of same post.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab1\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003ePost\u0026rsquo;s ID, features and trusties at level 0\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003ePost_ID\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eUser_ID\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eF0\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eF1\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eF2\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eF3\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eTrust_level_0\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.7\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.9\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.51\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.7\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.9\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.45\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.7\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.9\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.16\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.7\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.9\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.188\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e4\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.7\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.9\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.335\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eThis table shows the probability of validation, at the time of broadcasting, for 5 pieces of information with the same characteristics on the social network. For example, information 1, with the characteristics (0.7, 0.9, 0.2, 1), if broadcasted by user 0, will have a validity probability of 0.45.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ec. Trusties supervision of posts over the network\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTrusties\u0026rsquo; supervision under the developed environment consists of displaying how indicator post\u0026rsquo;s evolves over the time for each post.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eExample G(20,50)\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003eTo generate this graph (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e), this python code is executed:\u003c/p\u003e\n\u003cp\u003eG\u0026thinsp;=\u0026thinsp;nx.gnm_random_graph(nb_nodes\u0026thinsp;=\u0026thinsp;20, nb_edges\u0026thinsp;=\u0026thinsp;50)\u003c/p\u003e\n\u003cp\u003eOver the network of Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e, Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e shows the evolution of the probability of validity of each post, according to the level of its propagation in the social network.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab2\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003ePosts trusties at each level\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003edeep0\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003edeep1\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003edeep2\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003edeep3\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003epost1\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.202\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.2083125\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.2175\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.237928571\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003epost2\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.541\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.3803125\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.316944444\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.298871622\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003epost3\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.40225\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.379166667\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.3476625\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.337692308\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003epost4\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.466\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.416875\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.387096154\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.381377778\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003epost5\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.209\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.206975\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.207607143\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.209595745\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003epost6\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.10975\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.190958333\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.233590909\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.239518293\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003epost7\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.115\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.09705\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.109592105\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.117756579\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003epost8\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.12625\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.12375\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.112208333\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.104952703\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003epost9\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.09675\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.25734375\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.279407407\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.283244565\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003epost10\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.1035\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.233708333\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.268180556\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.287695946\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eThe graph of Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e clearly shows the evolution of this probability according to the spread of the posts in the social media. For example, post2 has a behavior of suspicious fake news broadcasted by a user who specializes in this news field, where its truth indicator quickly decreases, where its truth indicator quickly decreases.\u003c/p\u003e\n\u003cp\u003eOn the other hand, post10 behaves like a real post, broadcasted by a user who is not specialized in the content of this post. However, its probability (0.1035) at deep0 was visibly low, then it more than doubled at the next level (deep1), and continued to increase for the subsequent levels (deep2 and deep3). This can be justified by positive reactions of users at different levels (deep1, deep2, and deep3).\u003c/p\u003e\n\u003cp\u003eThe graph of Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e is another representation of the evolution of the post\u0026rsquo;s trusty\u0026rsquo;s indicator. In this graph, it can be noticed that \u003cem\u003epost1\u003c/em\u003e, \u003cem\u003epost5\u003c/em\u003e, \u003cem\u003epost7\u003c/em\u003e and post8 have certain stability over the social network. This allows to deduce that these posts can be trusted.\u003c/p\u003e\n\u003cp\u003eIndicators of trusty for certain posts such as post1, post6, post9, and post10, continue to increase as they spread further through the social network. This increase signifies positive backing from participants within the social network for these posts, consequently classifying them as genuine.\u003c/p\u003e\n\u003cp\u003eOn the other hand, trusties\u0026rsquo; factor for posts post2, post3, post4, and post8 continues to decrease as they propagate through the network. This indicates the adverse responses from social media users during the dissemination of these posts through the social network.\u003c/p\u003e"},{"header":"5. Conclusion and future works","content":"\u003cp\u003eThis work presented a new model for fake news detection in social media. For this purpose, it presented the overall behavior of a social media that can decipher fake news early as possible. This model represents the social capital by adequate concepts allowing adaptation in new technologies. The prediction mechanism is based on a probabilistic model, where users and their posts, over the social media, are modeled in this concept. The implicit supervision of the validity of a post in real-time and simulation results are promising. Future work will deal the implementation of this work over a real social network based on the IPFS distributed system.\u003c/p\u003e"},{"header":"Declarations","content":" \u003cp\u003eThe authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eLyazid HAMIMED wrote the main manuscript text, and all authors reviewed the manuscript.This work is a contribution of the PhD student Lyazid HAMIMED, supervised by Professor Mourad AMAD and co-supervised by Abdelmalek BOUDRIES.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eNyhan Brendan, Reifler Jason. \u003cem\u003eEstimating fact-checking\u0026rsquo;s effects\u003c/em\u003e, Arlington, VA: American Press Institute. 2015.\u003c/li\u003e\n\u003cli\u003eThorne James, Vlachos Andreas. \u003cem\u003eAutomated Fact Checking: Task Formulations, Methods and Future Directions\u003c/em\u003e, Proceedings of the 27th International Conference on Computational Linguistics. 2018. 3346\u0026ndash;3359.\u003c/li\u003e\n\u003cli\u003eThorne James, Vlachos Andreas, Christodoulopoulos Christos, Mittal Arpit. \u003cem\u003eFEVER: a Large-scale Dataset for Fact Extraction and VERification\u003c/em\u003e, Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018. 809\u0026ndash;819.\u003c/li\u003e\n\u003cli\u003eZellers Rowan, Holtzman Ari, Rashkin Hannah, Bisk Yonatan, Farhadi Ali, Roesner Franziska, Choi Yejin. \u003cem\u003eDefending Against Neural Fake News\u003c/em\u003e, arXiv preprint arXiv:1905.12616. 2019.\u003c/li\u003e\n\u003cli\u003eBaly Ramy, Karadzhov Georgi, Alexandrov Dimitar, Glass James, Nakov Preslav.\u003cem\u003e Predicting Factuality of Reporting and Bias of News Media Sources\u003c/em\u003e, Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Brussels, Belgium: Association for Computational Linguistics, X-XI 2018. 3528\u0026ndash;3539.\u003c/li\u003e\n\u003cli\u003eHashimoto Tatsunori, Zhang Hugh, Liang Percy. \u003cem\u003eUnifying Human and Statistical Evaluation for Natural Language Generation\u003c/em\u003e, Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). Minneapolis, Minnesota: Association for Computational Linguistics, VI 2019. 1689\u0026ndash;1701.\u003c/li\u003e\n\u003cli\u003eGehrmann Sebastian, Strobelt Hendrik, Rush Alexander. \u003cem\u003eGLTR: Statistical Detection and Visualization of Generated Text\u003c/em\u003e, Proceedings of the 57th Conference of the Association for Computational Linguistics: System Demonstrations. Florence, Italy: Association for Computational Linguistics, VII 2019. 111\u0026ndash;116.\u003c/li\u003e\n\u003cli\u003eBakhtin Anton, Gross Sam, Ott Myle, Deng Yuntian, Ranzato Marc\u0026rsquo;Aurelio, Szlam Arthur. \u003cem\u003eReal or Fake? Learning to Discriminate Machine from Human Generated Text\u003c/em\u003e, arXiv preprint arXiv:1906.03351. 2019.\u003c/li\u003e\n\u003cli\u003eP\u0026eacute;rez-Rosas Ver\u0026oacute;nica, Kleinberg Bennett, Lefevre Alexandra, Mihalcea Rada. \u003cem\u003eAutomatic Detection of Fake News\u003c/em\u003e, Proceedings of the 27th International Conference on Computational Linguistics. 2018. 3391\u0026ndash;3401.\u003c/li\u003e\n\u003cli\u003eVlachos Andreas, Riedel Sebastian. \u003cem\u003eFact Checking: Task definition and dataset construction\u003c/em\u003e, Proceedings of the ACL 2014 Workshop on Language Technologies and Computational Social Science. Baltimore, MD, USA: Association for Computational Linguistics, VI 2014. 18\u0026ndash;22.\u003c/li\u003e\n\u003cli\u003eWang William Yang. \u0026ldquo;\u003cem\u003eLiar, Liar Pants on Fire\u0026rdquo;: A New Benchmark Dataset for Fake News Detection\u003c/em\u003e, Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers). Vancouver, Canada: Association for Computational Linguistics, VII 2017. 422\u0026ndash;426.\u003c/li\u003e\n\u003cli\u003ePopat Kashyap, Mukherjee Subhabrata, Str\u0026ouml;tgen Jannik, Weikum Gerhard. \u003cem\u003eWhere the Truth Lies: Explaining the Credibility of Emerging Claims on the Web and Social\u003c/em\u003e Media, Proceedings of the 26th International Conference on World Wide Web Companion. Republic and Canton of Geneva, Switzerland: International World Wide Web Conferences Steering Committee, 2017. 1003\u0026ndash;1012. (WWW \u0026lsquo;17 Companion).\u003c/li\u003e\n\u003cli\u003eSchuster Tal, Shah Darsh J, Yeo Yun Jie Serene, Filizzola Daniel, Santus Enrico, Barzilay Regina. \u003cem\u003eTowards Debiasing Fact Verification Models\u003c/em\u003e, Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. 2019.\u003c/li\u003e\n\u003cli\u003e\u0026Aacute;. Figueira et L. Oliveira, \u0026laquo; \u003cem\u003eThe current state of fake news: challenges and opportunities\u003c/em\u003e \u0026raquo;, Procedia Computer Science, vol. 121, p. 817‑825, janv. 2017, doi: 10.1016/j.procs.2017.11.106.\u003c/li\u003e\n\u003cli\u003eH. Himdi, G. Weir, F. Assiri, et H. Al-Barhamtoshy, \u0026laquo; \u003cem\u003eArabic Fake News Detection Based on Textual Analysis\u003c/em\u003e \u0026raquo;, Arab J Sci Eng, vol. 47, no 8, p. 10453‑10469, ao\u0026ucirc;t 2022, doi: 10.1007/s13369-021-06449-y.\u003c/li\u003e\n\u003cli\u003eC. Busioc, S. Ruseti, et M. Dascalu, \u0026laquo; \u003cem\u003eA Literature Review of NLP Approaches to Fake News Detection and Their Applicability to Romanian-Language News Analysis\u003c/em\u003e \u0026raquo;, Transilvania, p. 65‑71, oct. 2020, doi: 10.51391/trva.2020.10.07.\u003c/li\u003e\n\u003cli\u003eS. C. Guntuku, D. B. Yaden, M. L. Kern, L. H. Ungar, et J. C. Eichstaedt, \u0026laquo; Detecting depression and mental illness on social media: an integrative review \u0026raquo;, Current Opinion in Behavioral Sciences, vol. 18, p. 43‑49, d\u0026eacute;c. 2017, doi: 10.1016/j.cobeha.2017.07.005.\u003c/li\u003e\n\u003cli\u003eM. Aldwairi et A. Alwahedi, \u0026laquo; Detecting Fake News in Social Media Networks \u0026raquo;, Procedia Computer Science, vol. 141, p. 215‑222, janv. 2018, doi: 10.1016/j.procs.2018.10.171.\u003c/li\u003e\n\u003cli\u003eM. A. Alonso, D. Vilares, C. G\u0026oacute;mez-Rodr\u0026iacute;guez, et J. Vilares, \u0026laquo; Sentiment Analysis for Fake News Detection \u0026raquo;, Electronics, vol. 10, no 11, Art. no 11, janv. 2021, doi: 10.3390/electronics10111348.\u003c/li\u003e\n\u003cli\u003eG. De Magistris, S. Russo, P. Roma, J. T. Starczewski, et C. Napoli, \u0026laquo; An Explainable Fake News Detector Based on Named Entity Recognition and Stance Classification Applied to COVID-19 \u0026raquo;, Information, vol. 13, no 3, Art. no 3, mars 2022, doi: 10.3390/info13030137.\u003c/li\u003e\n\u003cli\u003eX. Zhang et A. A. Ghorbani, \u0026laquo; An overview of online fake news: Characterization, detection, and discussion \u0026raquo;, Information Processing \u0026amp; Management, vol. 57, no 2, p. 102025, mars 2020, doi: 10.1016/j.ipm.2019.03.004.\u003c/li\u003e\n\u003cli\u003eK. Shu, H. R. Bernard, et H. Liu, \u0026laquo; Studying Fake News via Network Analysis: Detection and Mitigation \u0026raquo;, in Emerging Research Challenges and Opportunities in Computational Social Network Analysis and Mining, N. Agarwal, N. Dokoohaki, et S. Tokdemir, \u0026Eacute;d., in Lecture Notes in Social Networks. , Cham: Springer International Publishing, 2019, p. 43‑65. doi: 10.1007/978-3-319-94105-9_3.\u003c/li\u003e\n\u003cli\u003eA. Ard\u0026egrave;vol-Abreu, P. Delponti, et C. Rodr\u0026iacute;guez-Wang\u0026uuml;emert, \u0026laquo; Intentional or inadvertent fake news sharing? Fact-checking warnings and users\u0026rsquo; interaction with social media content \u0026raquo;, Profesional de la informaci\u0026oacute;n, vol. 29, no 5, Art. no 5, sept. 2020, doi: 10.3145/epi.2020.sep.07.\u003c/li\u003e\n\u003cli\u003e\u0026laquo; Fake News Detection Using Machine Learning and Deep Learning Algorithms \u0026raquo;. Consult\u0026eacute; le: 27 septembre 2023. [En ligne]. Disponible sur: https://ieeexplore.ieee.org/abstract/document/9436605/\u003c/li\u003e\n\u003cli\u003eZ. Chen et al., \u0026laquo; An Automatic Framework to Continuously Monitor Multi-Platform Information Spread \u0026raquo;, ao\u0026ucirc;t 2021.\u003c/li\u003e\n\u003cli\u003eK. Shu, L. Cui, S. Wang, D. Lee, et H. Liu, \u0026laquo; dEFEND: Explainable Fake News Detection \u0026raquo;, in Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery \u0026amp; Data Mining, in KDD \u0026rsquo;19. New York, NY, USA: Association for Computing Machinery, juill. 2019, p. 395‑405. doi: 10.1145/3292500.3330935.\u003c/li\u003e\n\u003cli\u003eS. Ahmed, K. Hinkelmann, et F. Corradini, \u0026laquo; Development of Fake News Model using Machine Learning through Natural Language Processing \u0026raquo;. arXiv, 19 janvier 2022. doi: 10.48550/arXiv.2201.07489.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Social networking, Social media, Fake news, Negative effect, social profiling, news features","lastPublishedDoi":"10.21203/rs.3.rs-4359937/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4359937/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eRecently, social media is becoming a stronger tool for spreading news in the world. These platforms make it easy for anyone to disseminate their ideas, flood the world by different types of information. In order to minimize the inconvenience of fake news inundation, most of the developed techniques aim at detecting fake news by exploring how they propagate on the social media. Minimizing the negative effect of this kind of information, needs stronger mechanisms to detect fake news at an early stage by focusing on their contents. This paper proposes a new model for trusties\u0026rsquo; content in social media. Its basic idea consists of combining news content and their propagation behavior over the social network. This model simulation shows that the susceptible fake news can be accused at an early stage. The performance evaluations show that the results are globally satisfactory.\u003c/p\u003e","manuscriptTitle":"A Collaborative and Real-Time Model for Trusties Content in Social Media","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-05-14 05:58:01","doi":"10.21203/rs.3.rs-4359937/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d4668435-5105-4acf-b780-73c9b307590e","owner":[],"postedDate":"May 14th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-05-14T05:58:04+00:00","versionOfRecord":[],"versionCreatedAt":"2024-05-14 05:58:01","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4359937","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4359937","identity":"rs-4359937","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}