Combating misinformation on social media platforms

Project Description

Social media platforms, such as Facebook, Twitter, YouTube, and Douyin, have become the primary information sources for billions of users worldwide. However, these platforms have been increasingly plagued by the widespread dissemination of misinformation, i.e., online content containing false information. The consequences of such misinformation are dire, particularly in critical domains like natural disasters, health, politics, and finance. This highlights the urgent need to address misinformation on social media platforms.

Existing efforts combating misinformation fall into two categories. First, human fact-checking, currently the primary approach employed by the major social media platforms, involves manual verification of content. Its process, while reliable, is inefficient and struggles to scale effectively given the massive volume of content generated on the platforms. Second, automatic fact-checking leverages advanced machine learning (ML) techniques, e.g., graph neural networks and large language models, to detect misinformation. The relevant approaches fall short due to their insufficient robustness in identifying misinformation with unseen characteristics across various aspects (such as content, topics, and linguistic styles as shown in Figure 1) or in the presence of various malicious attacks against the fact-checking methods. In short, misinformation on social media platforms has yet to be addressed by existing techniques.

With an aim to create a trustworthy digital future, this project has largely advanced the techniques for combating misinformation on social media platforms. To begin with, motivated by the lack of relevant data to train and benchmark robust ML-based solutions, we have constructed and proposed a large-scale and comprehensive dataset of fact-checking, featured by its multi-platform and multi-domain coverage of data. This publicly available dataset has attracted much attention in both academia and industry (see https://trustworthycomp.github.io/mcfend/). Besides, this project has thoroughly analysed the performance of a series of state-of-the-art misinformation detection methods in the presence of some feasible and ubiquitous attacks powered by large language models, which takes the first step to uncover the robustness of existing techniques and provide valuable insights for their improvements (see Figures 2 and 3).

In summary, this project, as an interdisciplinary effort that bridges social computing, artificial intelligence, and communication studies, delivers impactful contributions to combating misinformation on social media platforms by enhancing the robustness of fact-checking techniques, thus fostering a more trustworthy and resilient digital information ecosystem.

Fig 1. Examples of misinformation on different social media platforms with different characteristics across various aspects, such as content, topics, linguistic styles, etc.

Fig 2. An overview of our study regarding the robustness of existing methods of combating misinformation.

Fig 3. An illustration of existing state-of-the-art fact-checking techniques using graph neural networks and our designed novel attack method for analysing the robustness of these techniques.

Project Investigator

Dr Ivan LI (Department of Interactive Media)

Media Coverage

[Ming Pao 明報 2024] Media Interview and Coverage, “Training AI for Promoting Social Good”, 29/09/2024

• Summary: Dr LI Yupeng, a passionate interdisciplinary scholar is conducting interdisciplinary research in social computing, endeavouring to leverage the potential of Generated Artificial Intelligence (GenAI) to enhance various aspects of society.
• See also the online version: https://news.mingpao.com/pns/%e5%89%af%e5%88%8a/article/20240929/s00005/1727540340701   (in Chinese)

Data Analytics and Artificial Intelligence

EXPLORE MORE