A Latent Variable Model for Viewpoint Discovery from Threaded Forum Posts

Proceedings of NAACL-HLT 2013, pages 1031–1040, Atlanta, Georgia, 9–14 June 2013. c⃝2013 Association for Computational Linguistics A Latent Variable Model for Viewpoint Discovery from Threaded Forum Posts Minghui Qiu School of Information Systems Singapore Management University Singapore minghui.qiu.2010@smu.edu.sg Jing Jiang School of Information Systems Singapore Management University Singapore jingjiang@smu.edu.sg Abstract Threaded discussion forums provide an im- portant social media platform. Its rich user generated content has served as an important source of public feedback. To automatically discover the viewpoints or stances on hot is- sues from forum threads is an important and useful task. In this paper, we propose a novel latent variable model for viewpoint discov- ery from threaded forum posts. Our model is a principled generative latent variable model which captures three important factors: view- point specific topic preference, user identity and user interactions. Evaluation results show that our model clearly outperforms a number of baseline models in terms of both clustering posts based on viewpoints and clustering users with different viewpoints. 1 Introduction Threaded discussion forums provide an important social media platform that allows netizens to express their opinions, to ask for advice, and to form on- line communities. In particular, responses to major sociopolitical events and issues can often be found in discussion forums, which serve as an impor- tant source of public feedback. In such discussion threads, we often observe heated debates over a con- troversial issue, with different sides defending their viewpoints with different arguments. For example, after the presidential debate between Barack Obama and Mitt Romney, there were heated discussions in online forums like CreateDebate1 where some peo- ple expressed their support for Obama while some 1http://www.createdebate.com/ others have their opposition to him. For a user who is not closely following an event or issue, instead of going through all the existing posts in a long thread, she may want to quickly get an overview of the ma- jor viewpoints and arguments given by the different sides. For policy makers who want to obtain pub- lic feedback on social issues from social media, it is also desirable to automatically summarize the view- points on an issue from relevant threads. In this pa- per, we study the problem of modeling and discov- ering different viewpoints in forum threads. Recently there has been some work on finding contrastive viewpoints from text. The model pro- posed by Paul et al. (2010) assumes viewpoints and topics are orthogonal dimensions. Another model proposed by Fang et al. (2012) assumes that docu- ments are already grouped by viewpoints and it fo- cus on identifying contrastive viewpoint words un- der the same topic. However, these existing stud- ies are not based on interdependent documents like threaded forum posts. As a result, at least two im- portant characteristics of threaded forum data are not considered in these models. (1) User identity: The user or publisher of each forum post is known, and a user may publish several posts in the same thread. Since the same user’s opinion on an issue usually …

Content truncated to 500 words