Detecting Incongruent News Articles Using Multi-head Attention Dual Summarization

Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 967–977 November 20–23, 2022. ©2022 Association for Computational Linguistics 967 Detecting Incongruent News Articles Using Multi-head Attention Dual Summarization Sujit Kumar, Gaurav Kumar, Sanasam Ranbir Singh Department of Computer Science and Engineering Indian Institute of Technology, Guwahati, Assam, India {sujitkumar,ranbir}@iitg.ac.in, gauravchaudhary216@gmail.com Abstract With the increasing use of influencing incon- gruent news headlines for spreading fake news, detecting incongruent news articles has become an important research challenge. Most of the earlier studies on incongruity detection focus on estimating the similarity between the head- line and the encoding of the body or its sum- mary. However, most of these methods fail to handle incongruent news articles created with embedded noise. Motivated by the above is- sue, this paper proposes a Multi-head Attention Dual Summary (MADS) based method which generates two types of summaries that capture the congruent and incongruent parts in the body separately. From various experimental setups over three publicly available datasets, it is evi- dent that the proposed model outperforms the state-of-the-art baseline counterparts. 1 Introduction News headlines greatly influence opinion of the readers (Tannenbaum, 1953) and play a signifi- cant role in making a new viral on any social me- dia (Rieis et al., 2015) (Gabielkov et al., 2016) (Wei and Wan, 2017). A deceitful and incongruent news article can negatively affect readers, such as false beliefs and wrong opinions 12 (Ecker et al., 2014) (Ecker et al., 2022) (Tsfati et al., 2020). If a news headline misrepresents the content of its body then such headline and body pair is called in- congruent news article (Chesney et al., 2017) (Wei and Wan, 2017). In recent times, usage of decep- tive and incongruent news headlines as an effec- tive means to spread disinformation over digital platforms is evident (Chesney et al., 2017) (Ef- fron and Raj, 2020) 34. Consequently, detecting deceitful and incongruent news articles (Chesney et al., 2017) (Ecker et al., 2014) (Horner et al., 1Impact of misleading headline in health 2Misleading headlines effect on economy news 3Examples of misleading headline fake news 4Misleading headline fake news over WHO 2021) (Bago et al., 2020) (Guess et al., 2020) is be- coming an important research problem to counter the spread of misinformation over digital media. An incongruent news article may be constituted in various forms (i) the headline makes unrelated or opposite claims to its body, (ii) both headline and body refer to a common topic or event, but the contents are not related, (iii) both headline and body report a genuine event/incident, but the dates or name entities are manipulated, (iv) meth- ods are Earlier studies on incongruent news de- tection mainly focuses on estimating dissimilarity between headline and body using methods such as bag-of-words based features (Pomerleau and Rao, 2017), (Hanselowski et al., 2017), (Riedel et al., 2017), sequential encoding of headline and body (Hanselowski et al., 2018), …

Content truncated to 500 words