Mine the Easy, Classify the Hard: A Semi-Supervised Approach to Automatic Sentiment Classification

Proceedings of the 47th Annual Meeting of the ACL and the 4th IJCNLP of the AFNLP, pages 701–709, Suntec, Singapore, 2-7 August 2009. c⃝2009 ACL and AFNLP Mine the Easy, Classify the Hard: A Semi-Supervised Approach to Automatic Sentiment Classification Sajib Dasgupta and Vincent Ng Human Language Technology Research Institute University of Texas at Dallas Richardson, TX 75083-0688 {sajib,vince}@hlt.utdallas.edu Abstract Supervised polarity classification systems are typically domain-specific. Building these systems involves the expensive pro- cess of annotating a large amount of data for each domain. A potential solution to this corpus annotation bottleneck is to build unsupervised polarity classification systems. However, unsupervised learning of polarity is difficult, owing in part to the prevalence of sentimentally ambiguous re- views, where reviewers discuss both the positive and negative aspects of a prod- uct. To address this problem, we pro- pose a semi-supervised approach to senti- ment classification where we first mine the unambiguous reviews using spectral tech- niques and then exploit them to classify the ambiguous reviews via a novel com- bination of active learning, transductive learning, and ensemble learning. 1 Introduction Sentiment analysis has recently received a lot of attention in the Natural Language Processing (NLP) community. Polarity classification, whose goal is to determine whether the sentiment ex- pressed in a document is “thumbs up” or “thumbs down”, is arguably one of the most popular tasks in document-level sentiment analysis. Unlike topic-based text classification, where a high accu- racy can be achieved even for datasets with a large number of classes (e.g., 20 Newsgroups), polarity classification appears to be a more difficult task. One reason topic-based text classification is easier than polarity classification is that topic clusters are typically well-separated from each other, result- ing from the fact that word usage differs consid- erably between two topically-different documents. On the other hand, many reviews are sentimentally ambiguous for a variety of reasons. For instance, an author of a movie review may have negative opinions of the actors but at the same time talk enthusiastically about how much she enjoyed the plot. Here, the review is ambiguous because she discussed both the positive and negative aspects of the movie, which is not uncommon in reviews. As another example, a large portion of a movie re- view may be devoted exclusively to the plot, with the author only briefly expressing her sentiment at the end of the review. In this case, the review is ambiguous because the objective material in the review, which bears no sentiment orientation, sig- nificantly outnumbers its subjective counterpart. Realizing the challenges posed by ambiguous reviews, researchers have explored a number of techniques to improve supervised polarity classi- fiers. For instance, Pang and Lee (2004) train an independent subjectivity classifier to identify and remove objective sentences from a review prior to polarity classification. Koppel and Schler (2006) use neutral reviews to help improve the classi- fication of positive and negative reviews. More recently, McDonald et al. (2007) have investi- gated a model for jointly performing sentence- and document-level sentiment …

Content truncated to 500 words