Adapting a Polarity Lexicon using Integer Linear Programming for Domain-Specific Sentiment Classification

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 590–598, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Adapting a Polarity Lexicon using Integer Linear Programming for Domain-Specific Sentiment Classification Yejin Choi and Claire Cardie Department of Computer Science Cornell University Ithaca, NY 14853 {ychoi,cardie}@cs.cornell.edu Abstract Polarity lexicons have been a valuable re- source for sentiment analysis and opinion mining. There are a number of such lexi- cal resources available, but it is often sub- optimal to use them as is, because general purpose lexical resources do not reflect domain-specific lexical usage. In this pa- per, we propose a novel method based on integer linear programming that can adapt an existing lexicon into a new one to re- flect the characteristics of the data more directly. In particular, our method collec- tively considers the relations among words and opinion expressions to derive the most likely polarity of each lexical item (posi- tive, neutral, negative, or negator) for the given domain. Experimental results show that our lexicon adaptation technique im- proves the performance of fine-grained po- larity classification. 1 Introduction Polarity lexicons have been a valuable resource for sentiment analysis and opinion mining. In particu- lar, they have been an essential ingredient for fine- grained sentiment analysis (e.g., Kim and Hovy (2004), Kennedy and Inkpen (2005), Wilson et al. (2005)). Even though the polarity lexicon plays an important role (Section 3.1), it has received rela- tively less attention in previous research. In most cases, polarity lexicon construction is discussed only briefly as a preprocessing step for a sentiment analysis task (e.g., Hu and Liu (2004), Moilanen and Pulman (2007)), but the effect of different al- ternative polarity lexicons is not explicitly inves- tigated. Conversely, research efforts that focus on constructing a general purpose polarity lexicon (e.g., Takamura et al. (2005), Andreevskaia and Bergler (2006), Esuli and Sebastiani (2006), Rao and Ravichandran (2009)) generally evaluate the lexicon in isolation from any potentially relevant NLP task, and it is unclear how the new lexicon might affect end-to-end performance of a concrete NLP application. It might even be unrealistic to expect that there can be a general-purpose lexical resource that can be effective across all relevant NLP applica- tions, as general-purpose lexicons will not reflect domain-specific lexical usage. Indeed, Blitzer et al. (2007) note that the polarity of a particu- lar word can carry opposite sentiment depending on the domain (e.g., Andreevskaia and Bergler (2008)). In this paper, we propose a novel method based on integer linear programming to adapt an existing polarity lexicon into a new one to reflect the char- acteristics of the data more directly. In particular, our method considers the relations among words and opinion expressions collectively to derive the most likely polarity of each word for the given do- main. Figure 1 depicts the key insight of our approach using a bipartite graph. On the left hand side, each node represents a word, and on the right hand side, each node represents an opinion expression. …

Content truncated to 500 words