Structured Representation Learning for Online Debate Stance Prediction

Proceedings of the 27th International Conference on Computational Linguistics, pages 3728–3739 Santa Fe, New Mexico, USA, August 20-26, 2018. 3728 Structured Representation Learning for Online Debate Stance Prediction Chang Li Aldo Porco Dan Goldwasser Department of Computer Science Purdue University, West Lafayette, IN 47907 {li1873, aporco, dgoldwas}@purdue.edu Abstract Online debates can help provide valuable information about various perspectives on a wide range of issues. However, understanding the stances expressed in these debates is a highly challenging task, which requires modeling both textual content and users’ conversational interactions. Cur- rent approaches take a collective classification approach, which ignores the relationships between different debate topics. In this work, we suggest to view this task as a representation learning problem, and embed the text and authors jointly based on their interactions. We evaluate our model over the Internet Argumentation Corpus, and compare different approaches for structural information embedding. Experimental results show that our model can achieve significantly better results compared to previous competitive models. 1 Introduction In recent years, social media platforms play an increasingly important role in shaping political discourse. Online debate forums allow users to voice their opinions and engage with other users holding different views. Understanding the interactions between the users on these platforms can help provide insight into current political discourse, argumentation strategies and can help gauge public sentiment on policy issues on a large scale. The importance of understanding debate dialog has motivated significant research efforts (Somasundaran and Wiebe, 2010; Anand et al., 2011; Ghosh et al., 2014; Walker et al., 2012b; Hasan and Ng, 2013; Sridhar et al., 2015; Mohammad et al., 2016; Dong et al., 2017). In this paper, we focus on stance prediction, automatically identifying the stance expressed in debate posts on various issues. For example, Figure 1 describes a short debate dialog about marijuana legal- ization between three users (denoted a1, a2, a3). The content associated with each user is classified as supportive of legalization (PRO), or not (CON). “There are no deaths related to the actual use for marijuana this past year” “There are many things that are harmful. Alcohol is more harmful than marijuana. “But that doesn't mean marijuana should be made legal because the other two are.“ a2 a3 “Whether it kills people or not, it still is harmful to your body.” a2 a1 “If we were to make everything illegal because it was harmful we would be living with nothing.“ a3 Pro Con Pro Con Pro Author Discussion Text Stance Figure 1: Example of excerpts from a debate between three users about marijuana legalization. This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/. 3729 Early work took a text classification approach (Somasundaran and Wiebe, 2010; Anand et al., 2011), classifying individual posts using a rich feature set. Since debate posts are not written in isolation, but rather express the conversational interactions between users, modeling these interactions can help allevi- ate some of the difficulty of this task. More recent work take a collective …

Content truncated to 500 words