Project

Online social media platforms like Twitter and Facebook have been heavily used by many people for seeking, consuming, and sharing information. The openness and ease of interaction with those platforms from both desktop and handheld devices allow information to be propagated widely and rapidly at very low costs. This has both positive and negative aspects. On one hand, this helps people to quickly look for and follow stories and events of their interest, and thus be informed about important updates and useful lessons regarding these stories and events. On the other hand, this also enables the spreading of false information as social media users are generally more vulnerable and susceptible to such information. Among false information that is spreading on social media, fake news is the costliest one, both socially and economically. Fake news breaks the authenticity of the news ecosystems, thus trying to manipulate and persuade the news consumers to accept biases and false beliefs, which subsequently lead to seriously wrong decisions, e.g., vaccine resistance, or bad deals. A recent study estimated that fake news has been costing tens of billions USD yearly. Combating fake news on social media is therefore critical in any social sensing applications.

In this project, we aim to mitigate the above costs by addressing the identification of fake news in social media. Ideally, due to its prevalence, fake news should be detected automatically and as early as possible. However, this poses several challenges. Firstly, these news are often well written with the intention to mislead readers, and hence are not trivial to be detected purely based on their content. Secondly, the consumption and propagation of fake news by their consumers, which mainly distinguish them from the true ones, is not always observable due to privacy issues. Thirdly, as stories and events unfold, the truth value of news may also be changed overtime, which makes it even harder for evaluating the credibility of the news.

To overcome these challenges, we propose to develop a system that is able to collect and compile, from publicly available information sources, for each news propagated in social media, the evidence that is for or against the news. Our proposed system would then allow users to confidently and dynamically verify the news based on comparing and contrasting them with that extracted evidence. Furthermore, to deal with the multimodality of social media content, and to extract more convincing evidence, we aim to develop the system with capability of handling multimedia and cross-lingual data. For example, we would like to extract both images and news from multiple languages (e.g., English and Vietnamese) for supporting or rejecting a social media post, which may also contain both images and texts in multiple languages.

Designed as above, the proposed system is not only able to efficiently support users in detecting fake news but also able to provide transparent justification for any decision regarding the credibility of the news. This is a crucial feature that would make it be widely adopted and leveraged for social sensing applications. Potentially, the system can be easily integrated into any information system in organizations and institutes, and be useful for a wide range of non-technical users, including traders, journalists, and administrators.

Member

• Dr. NGUYEN Thi Minh Huyen, VNU University of Science, PI
• Dr. Tuan-Anh HOANG, VnTravel JSC, co-PI
• Dr. NGHIEM Thi Phuong, University of Science and Technology of Hanoi
• Dr. Thanh-Ha DO, VNU University of Science
• Dr. NGUYEN Luu Thuy Ngan, University of Information Technology, VNU-HCM
• Dr. TRAN Thi Oanh, VNU International School
• Dr. TRIEU Thanh Le, HCMC Univeristy of Social Sciences and Humanities
• Dr. NGUYEN Hai Vinh, VNU University of Science
• Prof. Ee-Peng Lim, Singapore Management University, advisor
• Prof. Seong G. Kong, Sejong University, advisor
• Mr. NGO The Quyen, VNU University of Science
• Mr. TO Tan Tai, VNU University of Science
• Mr. NGUYEN Duong Kien, VNU University of Science

Sponsors

Resource

• Hannotate: A flexible framework for text analytics
• More resource to be released. Stay tuned.

Publication

• Thanh-Nam Doan, Tuan-Anh Hoang; Benchmarking Neural Topic Models: An Empirical Study; Findings of the ACL; 2021
• Nguyen, Hoang H., Sergej Zerr, and Tuan-Anh Hoang; On Node Embedding of Uncertain Networks; 2020 IEEE International Conference on Big Data (Big Data). IEEE, 2020.