Sentence-Level Evidence Embedding for Claim Verification with Hierarchical Attention Networks

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2561–2571 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 2561 Sentence-Level Evidence Embedding for Claim Verification with Hierarchical Attention Networks Jing Ma1, Wei Gao2, Shafiq Joty3,4, Kam-Fai Wong1,5 1The Chinese University of Hong Kong, Hong Kong SAR 2Victoria University of Wellington, New Zealand 3Nanyang Technological University, Singapore 4Salesforce Research Asia, Singapore 5MoE Key Laboratory of High Confidence Software Technologies, China 1{majing,kfwong}@se.cuhk.edu.hk 2wei.gao@vuw.ac.nz, 3srjoty@ntu.edu.sg Abstract Claim verification is generally a task of ver- ifying the veracity of a given claim, which is critical to many downstream applications. It is cumbersome and inefficient for human fact-checkers to find consistent pieces of evi- dence, from which solid verdict could be in- ferred against the claim. In this paper, we pro- pose a novel end-to-end hierarchical attention network focusing on learning to represent co- herent evidence as well as their semantic relat- edness with the claim. Our model consists of three main components: 1) A coherence-based attention layer embeds coherent evidence con- sidering the claim and sentences from rele- vant articles; 2) An entailment-based attention layer attends on sentences that can semanti- cally infer the claim on top of the first atten- tion; and 3) An output layer predicts the ver- dict based on the embedded evidence. Ex- perimental results on three public benchmark datasets show that our proposed model outper- forms a set of state-of-the-art baselines. 1 Introduction The increasing popularity of social media has drastically changed how our daily news are pro- duced, disseminated and consumed.1 Without sys- tematic moderation, a large volume of informa- tion based on false or unverified claims (e.g., fake news, rumours, propagandas, etc.) can proliferate online. Such misinformation poses unprecedented challenges to information credibility, which tradi- tionally relies on fact-checkers to manually assess whether specific claims are true or not. Despite the increased demand, the effectiveness and efficiency of human fact-checking is handi- capped by the volume and fast pace the noteworthy 1The latest Pew Research statistics show that 68% American adults at least occasionally get news on social media. http://www.pewinternet.org/2018/03/ 01/social-media-use-in-2018/ claims being produced on daily basis. Therefore, it is an urgent need to automate the process and ease the human burden in assessing the veracity of claims (Thorne and Vlachos, 2018). Not surprisingly, various methods for automatic claim verification have been proposed using ma- chine learning. Typically, given the claims, mod- els are learned from auxiliary relevant sources such as news articles or social media responses for capturing words and linguistic units that might indicate viewpoint or language style towards the claim (Jin et al., 2016; Rashkin et al., 2017; Popat et al., 2017; Volkova et al., 2017; Dungs et al., 2018). However, the factuality of a claim is inde- pendent of people’s belief and subjective language use, and human perception is unconsciously prone to misinformation due to the common cognitive biases such as naive realism (Reed et al., 2013) and confirmation bias (Nickerson, 1998). A recent …

Content truncated to 500 words