A Classification-Based Approach for Implicit Feature Identification

Abstract

In recent years, sentiment analysis and opinion mining has grown to be one of the most active research areas. Most of the existing researches on feature-level opinion mining are dedicated to extract explicitly appeared features and opinion words. However, among the numerous kinds of reviews on the web, there are a significant number of reviews that contain only opinion words which imply some product features. The identification of such implicit features is still one of the most challenge tasks in opinion mining. In this paper, we propose a classification-based approach to deal with the task of implicit feature identification. Firstly, by exploiting the word segmentation, part-of-speech(POS) tagging and dependency parsing, a rule based method to extract the explicit feature-opinion pairs is presented. Secondly, the feature-opinion pairs for each opinion word are clustered and the training documents for each clustered feature-opinion pair are then constructed. Finally, the identification of implicit features is formulated into a classification-based feature selection. Experiments demonstrate that our approach outperforms the existing methods significantly.