Orientation Analysis for Feature-Dependent Opinion Words and Its Application in Online Comments

摘要

The analysis of sentiment orientation can be carried out at the document-level, sentence-level and feature-level. Detecting the orientation of the opinion words is the principal task for the feature-level sentiment classification. In real applications, the orientation of some opinion words relies on the features modified by the words, so there is dependency relationship existing between opinion words and features. This paper presents a novel analysis method of sentiment orientation for feature-dependent opinion words. We build the model of "opinion→feature" relationship based on statistics and text mining technology, and apply it on real world online comments. The experimental results show that the proposed model can detect the sentiment orientation of ambiguous words effectively and improve the F-measure of results compared to baseline.