The Unified Collocation Framework for Opinion Mining

Abstract

Opinion mining is a complicated text understanding technology involving opinion extraction and sentiment analysis. State-of-the-art techniques adopt idea of attribute-driven or sentiment-driven, leading to low opinion mining coverage. This paper proposes the unified collocation framework (UCF) and describes a novel unified collocation-driven (UCD) opinion mining method. The UCF incorporates attribute-sentiment collocations as well as their syntactical features to achieve reasonable generalization ability. Preliminary experiments show that 0.245 on averages improve recall of opinion extraction without obvious loss on opinion extraction precision and sentiment analysis accuracy.