From Argumentation Mining to Stance Classification

Proceedings of the 2nd Workshop on Argumentation Mining, pages 67–77, Denver, Colorado, June 4, 2015. c⃝2015 Association for Computational Linguistics From Argumentation Mining to Stance Classification Parinaz Sobhani1, Diana Inkpen1, Stan Matwin2 1 School of Electrical Engineering and Computer Science, University of Ottawa 2 Faculty of Computer Science, Dalhousie University 2 Institute of Computer Science, Polish Academy of Sciences psobh090@uottawa.ca, diana.Inkpen@uottawa.ca,stan@cs.dal.ca Abstract Argumentation mining and stance classifica- tion were recently introduced as interesting tasks in text mining. In this paper, a novel framework for argument tagging based on topic modeling is proposed. Unlike other ma- chine learning approaches for argument tag- ging which often require large set of labeled data, the proposed model is minimally super- vised and merely a one-to-one mapping be- tween the pre-defined argument set and the ex- tracted topics is required. These extracted ar- guments are subsequently exploited for stance classification. Additionally, a manually- annotated corpus for stance classification and argument tagging of online news comments is introduced and made available. Experi- ments on our collected corpus demonstrate the benefits of using topic-modeling for argument tagging. We show that using Non-Negative Matrix Factorization instead of Latent Dirich- let Allocation achieves better results for argu- ment classification, close to the results of a su- pervised classifier. Furthermore, the statistical model that leverages automatically-extracted arguments as features for stance classification shows promising results. 1 Introduction In the past, people were only the consumers of infor- mation on the web. With the advent of Web 2.0, new tools for producing User Generated Content (UGC) were provided. Consequently, huge amounts of text data is generate every day on the web. As the volume of this unstructured data increases, the request for automatically processing UGC grows significantly. Moreover, this new source of information and opin- ions contains valuable feedback about products, ser- vices, policies, and news and can play an important role in decision making for marketers, politicians, policy makers and even for ordinary people. So far, there has been a great effort toward sub- jectivity analysis of sentiment and opinion mining of reviews on concrete entities such as product or movies (Pang et al., 2002), (Dave et al., 2003), (Pang and Lee, 2005); however, this line of research does not fit online discussions opinion mining where comments not only contain the sentiment/stance of the commenter toward the target, but also convey personal beliefs about what is true or what action should be taken. This kind of subjectivity is called argumentation (Wilson and Wiebe, 2005). Argu- mentation analysis is more focused on the reason for author’s overall position. Stance has been defined as the overall position to- ward an idea, object or proposition (Somasundaran and Wiebe, 2010). There has been growing inter- est in stance classification particularly for online de- bates (Walker et al., 2012a), (Hasan and Ng, 2013). To the best of our knowledge, our paper is the first work for stance classification of the news comments considering particular news as target to investigate the overall position toward it. Argument …

Content truncated to 500 words