Discourse Connectors for Latent Subjectivity in Sentiment Analysis

Proceedings of NAACL-HLT 2013, pages 808–813, Atlanta, Georgia, 9–14 June 2013. c⃝2013 Association for Computational Linguistics Discourse Connectors for Latent Subjectivity in Sentiment Analysis Rakshit Trivedi College of Computing Georgia Institute of Technology Atlanta, GA 30308, USA rtrivedi6@gatech.edu Jacob Eisenstein School of Interactive Computing Georgia Institute of Technology Atlanta, GA 30308, USA jacobe@gatech.edu Abstract Document-level sentiment analysis can bene- fit from fine-grained subjectivity, so that sen- timent polarity judgments are based on the relevant parts of the document. While fine- grained subjectivity annotations are rarely available, encouraging results have been ob- tained by modeling subjectivity as a latent variable. However, latent variable models fail to capitalize on our linguistic knowledge about discourse structure. We present a new method for injecting linguistic knowledge into latent variable subjectivity modeling, using discourse connectors. Connector-augmented transition features allow the latent variable model to learn the relevance of discourse con- nectors for subjectivity transitions, without subjectivity annotations. This yields signif- icantly improved performance on document- level sentiment analysis in English and Span- ish. We also describe a simple heuristic for automatically identifying connectors when no predefined list is available. 1 Introduction Document-level sentiment analysis can benefit from consideration of discourse structure. Voll and Taboada (2007) show that adjective-based sentiment classification is improved by examining topicality (whether each sentence is central to the overall point); Yessenalina et al. (2010b) show that bag-of- ngrams sentiment classification is improved by ex- amining subjectivity (whether a sentence expresses a subjective opinion or objective fact). However, it is unclear how best to obtain the appropriate dis- course analyses. Voll and Taboada (2007) find that domain-independent discourse parsing (Soricut and Marcu, 2003) offers little improvement for senti- ment analysis, so they resort to training a domain- specific model for identifying topic sentences in re- views. But this requires a labeled dataset of topic sentences, imposing a substantial additional cost. Yessenalina et al. (2010b) treat sentence level subjectivity as a latent variable, automatically in- ducing the “annotator rationale” (Zaidan et al., 2007; Yessenalina et al., 2010a) for each training sen- tence so as to focus sentiment learning on the sub- jective parts of the document. This yields sig- nificant improvements over bag-of-ngrams super- vised sentiment classification. Latent variable sub- jectivity analysis is attractive because it requires neither subjectivity annotations nor an accurate domain-independent discourse parser. But while the “knowledge-free” nature of this approach is appeal- ing, it is unsatisfying that it fails to exploit decades of research on discourse structure. In this paper, we explore a lightweight approach to injecting linguistic knowledge into latent variable models of subjectivity. The entry point is a set of discourse connectors: words and phrases that signal a shift or continuation in the discourse structure. Such connectors have been the subject of exten- sive study in the creation of the Penn Discourse Treebank (PDTB: Prasad et al. 2008). The role of discourse connectors in sentiment analysis can be clearly seen in examples, such as “It’s hard to imagine the studios hiring another manic German maverick to helm …

Content truncated to 500 words