Automatic Construction of Domain-specific Sentiment Lexicon for Context-Sensitive Opinion Retrieval

Existing opinion retrieval systems often employ manually constructed generic sentimentlexicons to identify sentiment words embedded in opinionated texts such as product reviews orblog posts. However, manually constructing sentiment lexicons is very time-consuming and it may notbe feasible for certain application domains where annotation expertise is not available.The main contribution of this paper is thedevelopment of a novel Kullback-Leibler divergence based computational method for the automaticconstruction of domain-specific sentimentlexicons to enhance cross-domain opinion retrieval.Our initial experiments show that theproposed methodology can automatically generate domain-specific sentiment lexicons whichcontribute to improve the effectiveness of opinion retrieval at the document level. In particular, for thesentiment polarity detection task, an average improvement of F-measure by $6.56 \%$is achieved when compared to the performance of a baseline system which employsa widely used generic sentiment lexicon. Our research opens the door to the developmentof more effective opinion retrieval system to extract business intelligence from hugenumber opinionated expressions posted to the Web.