Article Reranking by Memory-Enhanced Key Sentence Matching for Detecting Previously Fact-Checked Claims

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 5468–5481 August 1–6, 2021. ©2021 Association for Computational Linguistics 5468 Article Reranking by Memory-Enhanced Key Sentence Matching for Detecting Previously Fact-Checked Claims Qiang Sheng1,2, Juan Cao1,2, Xueyao Zhang1,2, Xirong Li3, Lei Zhong1,2 1 Key Laboratory of Intelligent Information Processing, Institute of Computing Technology, Chinese Academy of Sciences 2University of Chinese Academy of Sciences 3Key Lab of Data Engineering and Knowledge Engineering, Renmin University of China {shengqiang18z,caojuan,zhangxueyao19s,zhonglei18s}@ict.ac.cn xirong@ruc.edu.cn Abstract False claims that have been previously fact- checked can still spread on social media. To mitigate their continual spread, detecting previously fact-checked claims is indispens- able. Given a claim, existing works retrieve fact-checking articles (FC-articles) for detec- tion and focus on reranking candidate arti- cles in the typical two-stage retrieval frame- work. However, their performance may be limited as they ignore the following charac- teristics of FC-articles: (1) claims are often quoted to describe the checked events, pro- viding lexical information besides semantics; and (2) sentence templates to introduce or debunk claims are common across articles, providing pattern information. In this paper, we propose a novel reranker, MTM (Memory- enhanced Transformers for Matching), to rank FC-articles using key sentences selected us- ing event (lexical and semantic) and pattern information. For event information, we pro- pose to finetune the Transformer with regres- sion of ROUGE. For pattern information, we generate pattern vectors as a memory bank to match with the parts containing patterns. By fusing event and pattern information, we se- lect key sentences to represent an article and then predict if the article fact-checks the given claim using the claim, key sentences, and pat- terns. Experiments on two real-world datasets show that MTM outperforms existing methods. Human evaluation proves that MTM can cap- ture key sentences for explanations. The code and the dataset are at https://github.com/ ICTMCG/MTM. 1 Introduction Social media posts with false claims have led to real-world threats on many aspects such as pol- itics (Fisher et al., 2016), social order (Wang and Li, 2011), and personal health (Chen, 2020). Hot lemonade can kill cancer cells without hurting normal cells. Stage1 BM25 Sentences in Candidate Fact-checking Articles Sentence Relevant? Contains Quotation? Contains Fact-checking Patterns? S1. Lemon is not so-called acid food, and drinking lemonade does not lead to cancer. (From Article 1) No, but on a similar topic No No S2. The rumor saying hot lemonade can kill cancer cells has spread over years. (From Article 2) Yes Yes (Underlined) Yes (In boldface) S3. It is just a groundless inference that lemon has a curative effect of cancer. (From Article 3) Yes No Yes (In boldface) Claim Fact-checking Articles Stage2 MTM Manual Display to Fact-checkers Automatic Make rules based on relevance scores The scope of our work Claim Ranked Candidate Articles Determine if a claim has been fact-checked Candidate Fact-checking Articles (a) (b) Figure 1: (a) Workflow of detecting a previously fact- checked claim. Our model MTM …

Content truncated to 500 words