Mining for Opinions Across Domains: A Cross-Language Study

P. 67–74.

Kravchenko A.

An important task in opinion mining is detecting subjective expressions in texts and distinguishing them from factual information. High lexicon diversity between different domains excludes the possibility of formulating universal rules that would work for any area of knowledge.  In this article we suggest a solution for this problem. We define the features that most opinionated sentences share and propose a cross-language classification of subjective expressions, illustrated by examples in Russian, English and Chinese. We also propose an algorithm based on this classification that generates a set of extraction patterns for any domain from a corpus of untagged texts. The corpus requires no additional preparation except for POS-tagging.  The effectiveness of the proposed approach is evaluated for English and Russian on collections of approximately 300 000 sentences each, gathered from three different domains: user reviews on movies, headphones and photo cameras.

 

Keywords: sentiment analysis, opinion mining, cross-language information retrieval

In book

Proceedings of the First International Workshop on Sentiment Discovery from Affective Data (SDAD 2012), Leuven, Belgium, 2012

Leuven: [б.и.], 2012.