KCD: Knowledge Walks and Textual Cues Enhanced Political Perspective Detection in News Media

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4129 - 4140 July 10-15, 2022 ©2022 Association for Computational Linguistics KCD: Knowledge Walks and Textual Cues Enhanced Political Perspective Detection in News Media Wenqian Zhang♠∗ Shangbin Feng♠∗ Zilong Chen♢∗ Zhenyu Lei♢ Jundong Li♡ Minnan Luo♠† School of Computer Science and Technology, Xi’an Jiaotong University♠ School of Science, Xi’an Jiaotong University♢, University of Virginia♡ {2194510944,wind_binteng,luoyangczl,Fischer}@stu.xjtu.edu.cn jundong@virginia.edu minnluo@xjtu.edu.cn Abstract Political perspective detection has become an increasingly important task that can help com- bat echo chambers and political polarization. Previous approaches generally focus on lever- aging textual content to identify stances, while they fail to reason with background knowledge or leverage the rich semantic and syntactic tex- tual labels in news articles. In light of these limitations, we propose KCD, a political per- spective detection approach to enable multi- hop knowledge reasoning and incorporate tex- tual cues as paragraph-level labels. Specifi- cally, we firstly generate random walks on ex- ternal knowledge graphs and infuse them with news text representations. We then construct a heterogeneous information network to jointly model news content as well as semantic, syn- tactic and entity cues in news articles. Finally, we adopt relational graph neural networks for graph-level representation learning and conduct political perspective detection. Extensive exper- iments demonstrate that our approach outper- forms state-of-the-art methods on two bench- mark datasets. We further examine the effect of knowledge walks and textual cues and how they contribute to our approach’s data efficiency. 1 Introduction Political perspective detection aims to identify ide- ological stances of textual data such as social me- dia posts and news articles. Previous approaches generally leverage the textual content of news arti- cles with various text modeling techniques to iden- tify political stances. Those works (Jiang et al., 2019; Li and Goldwasser, 2019, 2021; Feng et al., 2021a) leveraged diversified text models, such as recurrent neural networks (Yang et al., 2016), word embedding techniques (Pennington et al., 2014; Pe- ters et al., 2018), convolutional neural networks (Jiang et al., 2019), and pre-trained language mod- els (Devlin et al., 2019; Liu et al., 2019), to encode ∗These authors contributed equally to this work. †Corresponding author. Source: Daily Kos Stance: Left …… CNN is reporting that the Trump campaign were offered access to Wikileaks documents, including special access to a Wikileaks website, a month before Wikileaks began publishing …… CNN is reporting that the Trump campaign were offered access to Wikileaks documents, including special access to a Wikileaks website, a month before Wikileaks began publishing …… The email, which apparently slipped the memory of history’s most forgetful campaign team, included a decryption key and address for documents stolen by Russian hackers and later …… He claimed that …… “…… That means that the vast amount of stuff that we are publishing about Clinton will have a much higher impact, because it won’t be perceived as coming from ……” NewYork President BillClinton DonaldTrump IsA IsA (DonaldTrump, RunFor, 2016U.S.PresidentialElection, RunFor, …

Content truncated to 500 words