Harnessing Abstractive Summarization for Fact-Checked Claim Detection

Proceedings of the 29th International Conference on Computational Linguistics, pages 2934–2945 October 12–17, 2022. 2934 Harnessing Abstractive Summarization for Fact-Checked Claim Detection Varad Bhatnagar1, Diptesh Kanojia2,3, Kameswari Chebrolu1 1Department of Computer Science and Engineering, IIT Bombay, India 2Surrey Institute for People-Centred AI, 3Department of Computer Science 2,3University of Surrey, United Kingdom 1{varadhbhatnagar,chebrolu}@cse.iitb.ac.in 2,3d.kanojia@surrey.ac.uk Abstract Social media platforms have become new bat- tlegrounds for anti-social elements, with mis- information being the weapon of choice. Fact- checking organizations try to debunk as many claims as possible while staying true to their journalistic processes but cannot cope with its rapid dissemination. We believe that the so- lution lies in partial automation of the fact- checking life cycle, saving human time for tasks which require high cognition. We propose a new workflow for efficiently detecting previ- ously fact-checked claims that uses abstractive summarization to generate crisp queries. These queries can then be executed on a general- purpose retrieval system associated with a col- lection of previously fact-checked claims. We curate an abstractive text summarization dataset comprising noisy claims from Twitter and their gold summaries. It is shown that retrieval per- formance improves 2x by using popular out- of-the-box summarization models and 3x by fine-tuning them on the accompanying dataset compared to verbatim querying. Our approach achieves Recall@5 and MRR of 35% and 0.3, compared to baseline values of 10% and 0.1, respectively. Our dataset, code, and models are available publicly here. 1 Introduction Social media is increasingly used for business, en- tertainment, and political discourse, thus, encour- aging users to produce and consume large volumes of information that may not always be accurate. Due to a lack of digital awareness, the masses often believe and forward such disputed claims in their social circles. Such spread of misinformation often culminates in incidents which cause damage to life and property. It is well documented that misinfor- mation is used as a tool by political agents to slan- der their opposition (Allcott and Gentzkow, 2017) and influence the opinion of the masses. It becomes furthermore dangerous when such claims pertain to religious beliefs, often leading to violence and mob lynchings1. In the era of COVID-19, unverified medical advice has also been circulated on social media (Shahi et al., 2021) which has already led to various health hazards. Social media platforms have undertaken con- certed efforts to tackle the fake news epidemic by enforcing strict policies to weed out unverified and sensitive content and ban habitual offenders. Jour- nalistic organizations such as Alt News2, Factly3, Boom Live4 and Snopes5 among others are also fighting this problem by publishing fact-checking articles investigating the veracity of viral dubious claims. These articles detail the journalistic proce- dures followed to fact-check the claim along with suitable references. Numerous researchers are working on AI- based solutions for fact-checking claims. Many datasets (Thorne et al., 2018; Sathe et al., 2020; Fan et al., 2020; Schuster et al., 2021) have been released to train models which can automate sub- tasks such as claim verification, evidence …

Content truncated to 500 words