Accessibility statement

Fake news detection, a deep learning approach

Social media has quickly become the main way many in the world get their news; this has democratised the way information is shared, and therefore the amount of “news” content online has exploded. This large volume of “news” content has naturally increased the amount of unreliable and low-quality information that is seen and engaged with. If left unchecked this information has the potential to do a huge amount of harm, even if only allowed to spread for a short period of time.

Due to the huge volumes and rapid spread of digital information, it is impossible for human moderators and fact checkers to keep up. This has motivated research into automatic fake news detection. Some automatic detection methods have attempted to use headlines and article text to detect fake news while others have attempted to utilise social context information. The performance of the existing methods including machine learning and deep learning models, however, remains unsatisfactory, largely due to, for example, inadequate amount of evidence, inadequate reasoning, or inadequate consideration of other reasons for why an argument is convincing.

The project seeks to explore the use of argumentation technology to facilitate better fake news detection. Typically, the project investigates the use of Toulmin’s argument model (i.e. warrant, rebuttals as shown in Figure 1) to support the reasoning for the veracity of a claim.

Figure 1: An example of Toulmin’s argument model reproduced from wikipedia.

The project involves the use of deep learning models and natural language processing techniques to summarise and generate arguments for use by the task of stance detection and veracity checking. Argument summarisation and generation falls within the area of argument mining, and the use of which in the IBM debater has received a lot of media attention.

People involved:

Papers:

F. T. Al-Khawaldeh, T. Yuan and D. Kazakov , ‘A Novel Model for Enhancing Fact-Checking’, in Proceedings of the Computing Conference, London, July 2021.

F. T. Al-Khawaldeh, T. Yuan and D. Kazakov, ‘RL-GAN Based Toulmin Argument’, Journal of Applied Science and Computations (JASC), vol. VII, no. III, pp. 106–120, 2020.

F. T. Al-Khawaldeh, T. Yuan and D. Kazakov,, ‘Linguistic Style-Aware Hybrid Model for Cross-Domain Factuality Checking’, Journal of Applied Science and Computations (JASC), vol. VII, no. IV, pp. 7–20, 2020.

F. T. Al-Khawaldeh, T. Yuan and D. Kazakov, ‘Integrating Stance Detection and Factuality Checking’, International Journal of Advanced Studies in Computer Science and Engineering (IJASCSE), vol. 9, no. 3, pp. 1–17, 2020.