Finding Thoughtful Comments from Social Media

Proceedings of COLING 2012: Technical Papers, pages 995–1010, COLING 2012, Mumbai, December 2012. Finding Thoughtful Comments from Social Media Swapna Gottipati, Jing Jiang School of Information Systems, Singapore Management University, Singapore swapnag.2010@smu.edu.sg, jingjiang@smu.edu.sg ABSTRACT Online user comments contain valuable user opinions. Comments vary greatly in quality and detecting high quality comments is a subtask of opinion mining and summarization research. Finding attentive comments that provide some reasoning is highly valuable in understanding the user’s opinion particularly in sociopolitical opinion mining and aids policy makers, social organizations or government sectors in decision making. In this paper we study the problem of detecting thoughtful comments. We empirically study various textual features, discourse relations and relevance features to predict thoughtful comments. We use logistic regression model and test on the datasets related to sociopolitical content. We found that the most useful features include the discourse relations and relevance features along with basic textual features to predict the comment quality in terms of thoughtfulness. In our experiments on two different datasets, we could achieve a prediction score of 79.37% and 73.47% in terms of F-measure on the two data sets, respectively. KEYWORDS: Opinion mining, Information Extraction, Text Classification. 995 1 Introduction In recent years sentiment analysis and opinion mining has been extensively studied in natural language processing (Pang and Lee, 2008), largely because of the availability of a huge amount of opinionated text in online product reviews, blogs, social networking sites, forums, etc. Most work on opinion mining is about mining reviews of products and services, and the focus of these studies has been on a few important subtasks such as sentiment classification (Pang et al., 2002; Cui et al., 2006; Jiang et al., 2011) and opinion extraction (Popescu and Etzioni, 2005; Choi et al., 2006; Wu et al., 2009). When we go beyond product review mining and consider the general problem of opinion mining from social media, many other subtasks and challenges arise. One of them is how to assess the quality of online comments and select high quality ones for further analysis and summarization. Consider the problem of mining the comments found in online social media towards a political speech such as Obama’s State of the Union address. By restricting the search space to politically active blogs and forums and by using queries such as “State of the Union,” likely we are able to retrieve highly relevant comments to the speech. However, not every comment contains valuable insight into the public’s opinions regarding the sociopolitical issues addressed in the speech. Comments such as “innovate and innovation appeared 10 times” and “To him..investment = more deficit spending” are subjective but lack thoughtful explanations to support their claims. In comparison, comments like “You want to really drive innovation, job growth and entrepreneurs? Make education, health care and retirement less of a burden on the average family, adopt more socialist policies like Norway (paid for by higher taxes, especially on the rich), and watch our standard of living rise at last!” provide much more insightful …

Content truncated to 500 words