Construction of a Model for the Cross-Domain Opinion Word Extraction

In this paper we consider a new approach for domain-specific opinion word extraction in the Russian language. We propose a set of statistical features and an algorithm combination that can extract opinion words in a particular domain. The extraction model was trained in the movie domain and then applied to four other domains. The quality of the obtained sentiment lexicons was evaluated intrinsically on the base of an expert markup and remained on the high level during the model transfer to various domains. Finally, our method is adapted to the movie domain in English and it demonstrated good results.