Joint Models of Disagreement and Stance in Online Debate

Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, pages 116–125, Beijing, China, July 26-31, 2015. c⃝2015 Association for Computational Linguistics Joint Models of Disagreement and Stance in Online Debate Dhanya Sridhar,1 James Foulds,1 Bert Huang,2 Lise Getoor,1 Marilyn Walker1 1Department of Computer Science, University of California Santa Cruz {dsridhar, jfoulds, getoor, mawalker}@ucsc.edu 2Department of Computer Science, Virginia Tech bhuang@vt.edu Abstract Online debate forums present a valu- able opportunity for the understanding and modeling of dialogue. To understand these debates, a key challenge is inferring the stances of the participants, all of which are interrelated and dependent. While collectively modeling users’ stances has been shown to be effective (Walker et al., 2012c; Hasan and Ng, 2013), there are many modeling decisions whose ramifi- cations are not well understood. To in- vestigate these choices and their effects, we introduce a scalable unified probabilis- tic modeling framework for stance clas- sification models that 1) are collective, 2) reason about disagreement, and 3) can model stance at either the author level or at the post level. We comprehensively evaluate the possible modeling choices on eight topics across two online debate cor- pora, finding accuracy improvements of up to 11.5 percentage points over a local classifier. Our results highlight the im- portance of making the correct modeling choices for online dialogues, and having a unified probabilistic modeling framework that makes this possible. 1 Introduction Understanding stance and opinion in dialogues can provide critical insight into the theoretical un- derpinnings of discourse, argumentation, and sen- timent. Systems for predicting the stances of indi- viduals can potentially have positive social impact and are of practical interest to non-profits, govern- mental organizations, and companies. For exam- Dialogue Turns Stance User 1: 18. That’s the smoking age thats the shooting age. Why do you think they call it ATF? ANTI User 2: Shooting age? I know 7 year old shooters. 18 should be the gun purchasing age, but there is really no ”shooting” age. ANTI User 1: I know. I was just pointing out that the logic used to propose a 21 year ”shooting age” was inconsistent. ANTI User 2: I see. I dont think its really fair that you can join the army at 18 and use handguns and military weapons, but you cant purchase a handgun until 21. ANTI Figure 1: Example of a debate dialogue turn be- tween two users on the gun control topic, from 4FORUMS.COM. ple, stance predictions may be used to target pub- lic awareness and advocacy campaigns, direct po- litical fundraising and get-out-the vote efforts, and improve personalized recommendations. Online debate websites are a particularly rich source of argumentative dialogic data (Fig. 1). On these websites, users debate and share their opin- ions on a variety of social and political issues. Previous work (Somasundaran and Wiebe, 2010; Walker et al., 2012c) has shown that stance clas- sification in online debates is a challenging prob- lem. While collective approaches …

Content truncated to 500 words