Fine-Grained Product Feature Extraction in Chinese Reviews

Abstract

Fine-grained product feature extraction is the most important task in opinion mining. To realize the fine-grained product feature extraction in Chinese reviews, three main tasks have been solved in this paper. Firstly, we propose a dependency parsing based method to directly extract the explicit feature-opinion pairs. Then, by analyzing the characteristics of two synonyms features and the relations with opinion words, we calculate the similarities to cluster features. Finally, we propose a novel implicit feature extraction method by combining review context information and two kind opinions to extract implicit features. Experiments show that the dependency parsing based method can get high precision, by considering verbs as product feature can improve the recall obviously. Besides, several proven pruning strategies can improve the accuracy. The comparison demonstrates that our implicit feature extraction method outperforms existing method, and feature clustering before implicit feature mining can get better results.