The Battle Against Fake News

Dr. Jing Ma

Recent years have witnessed the explosion of online misinformation (fake news, rumours, disinformation, etc), which have become an increasingly daunting issue in society. With an aim to seek to tackle the issues, Dr. Jing Ma, Department of Computer Science and System Health Lab, HKBU, dealt with multiplatform fact verification and rumour detection models in her two recent publications.

Multiplatform Fact and Language Verification

Fact Verification identifies the veracity of an event or a topic. Existing methods for fact verification essentially focus on recognising the majority opinions expressed towards the respective posts by capturing rumour-indicative signals from a large-scale corpus. However, this becomes more challenging and complex when the event is simultaneously propagated on multiple platforms or media outlets, and the languages used to report the event are different. Challenges include:

  1. The properties of the online information (e.g., users’ opinion, language style, propagated on multiplatforms can be vastly diverse)
  2. The language/knowledge gaps can result in the distortion of factual statements via mistranslation or misunderstanding
  3. There are no prior works or datasets that attempt to study multiplatforms and multi-languages in the area of fact verification

Dr Ma’s team has developed a new approach for reasoning evidence derived from multiple platforms with the aim of inferring the truthfulness of a given event. For example, the set of documents or web snippets from different media outlets (i.e., platforms) reporting the same (fake) event about “smoking, methanol or cocaine can cure for the coronavirus” would form a storyline. Such storylines are automatically discovered and constructed via comparing and matching among the related contents. The capability of conducting comparisons among different platforms will have important, positive effects on the quality and availability of online information, helping to detect inconsistencies and missing content, thus fighting against misinformation campaigns.

 

Fig 1
A Figure from Dr Ma’s recently published study “Debunking Rumors on Twitter with Tree Transformer.” Source: https://www.aclweb.org/anthology/2020.coling-main.476/

 

Furthermore, inaccurate translation from one language to another can bring about misinformation. Dr Ma’s team is developing a translation system to further improve the flow of information across different languages, allowing them to address language gaps to help human fact-checkers to early detect the presence of suspicious content. “An important part of this research is assessing bilingual fake news about the COVID-19 pandemic,” Dr Ma asserted. “This is particularly important in a large bilingual setting such as Hong Kong.”

Rumour Detection Model

As previously shown, rumours spreading on social media severely jeopardises the credibility of online content. Thus, the automatic debunking of rumours is of great importance to keep social media a healthy environment. While facing a dubious claim, people often dispute its truthfulness in their posts containing various cues, which can form useful evidence with long-distance dependencies. Dr Ma’s team sought to learn discriminative features from microblog posts by following their non-sequential propagation structure and generating more powerful representations for identifying rumours. They reveal that effective rumour detection is highly related to finding evidential posts, essentially the posts expressing a specific attitude towards the veracity of a claim. The results confirm that:

  1. The models achieve much better rumour detection and classification performance than state-of-the-art approaches
  2. The attention mechanisms for focusing on evidential posts can further improve the performance of the team’s method
  3. The approach possesses superior capacity on detecting rumours at the very early stage.

 

The full findings can be found here.

 

Dr Ma received her PhD degree from the Department of Systems Engineering and Engineering Management, The Chinese University of Hong Kong in 2020. Before that, she obtained both her B.E. and M.E. degrees from Beijing University of Posts and Telecommunications in 2013 and 2016 respectively. Her research includes Natural Language Processing, Social Network Analysis and Mining, Rumor Detection and Fact Verification.

 

Related Publication

Source: Research Office