Predicting Stances from Social Media Posts using Factorization Machines

Proceedings of the 27th International Conference on Computational Linguistics, pages 3381–3390 Santa Fe, New Mexico, USA, August 20-26, 2018. 3381 Predicting Stances from Social Media Posts using Factorization Machines Akira Sasaki1∗, Kazuaki Hanawa2, Naoaki Okazaki3,4, Kentaro Inui2,5 1Recruit Technologies Co., Ltd., Tokyo, Japan 2Graduate School of Information Sciences, Tohoku University, Miyagi, Japan 3School of Computing, Tokyo Institute of Technology, Tokyo, Japan 4National Institute of Advanced Industrial Science and Technology (AIST), Tokyo, Japan 5RIKEN Center for Advanced Intelligence Project (AIP), Tokyo, Japan 1sakira@r.recruit.co.jp 2{hanawa,inui}@ecei.tohoku.ac.jp 3okazaki@c.titech.ac.jp Abstract Social media provide platforms to express, discuss, and shape opinions about events and issues in the real world. An important step to analyze the discussions on social media and to assist in healthy decision-making is stance detection. This paper presents an approach to detect the stance of a user toward a topic based on their stances toward other topics and the social media posts of the user. We apply factorization machines, a widely used method in item recommendation, to model user preferences toward topics from the social media data. The experimental results demonstrate that users’ posts are useful to model topic preferences and therefore predict stances of silent users. 1 Introduction Web and social media provide platforms to express, discuss, and shape opinions about events and issues in the real world. However, these platforms suffer from new emerging problems such as filter bub- ble (Pariser, 2011), echo chamber (Jamieson and Cappella, 2008), and fake news (Lazer et al., 2018). An important step to analyze the argumentation structure of social media and to assist in healthy decision- making is stance detection (Mohammad et al., 2016): predicting whether a given text/user is in favor (agree), against (disagree), or neutral toward a target topic (e.g., Donald Trump). Unfortunately, users rarely make an explicit statement about a topic. For example, computers may easily detect a stance for the topic of the Trans-Pacific Partnership (TPP) from the sentence, “I totally disagree with TPP,” containing the topic word ‘TPP’ with the explicit linguistic pattern “I totally disagree with ...” However, many social media posts may not refer to a topic but only to its related topics, for example, “We should protect Japanese agriculture.” We need commonsense knowledge about the topic (e.g., “TPP suppresses Japanese agriculture”) to predict the stance (objection to TPP) for such a sentence (see Figure 1). Furthermore, social media users are often silent on a topic (Gong et al., 2016): users who are interested in a topic without posting anything related to it are called the silent majority (Lassiter, 2007) and lurkers (Gong et al., 2015). This paper presents an approach to detect the stance of a user toward a topic based on: (1) their stances toward other topics, and (2) the social media posts of the user. We apply factorization machines (FMs), a widely used method in item recommendation, to model user preferences toward topics from data consisting of the stance statements and social media posts of users. Contributions of this work are twofold. 1. We present …

Content truncated to 500 words