Product Feature Extraction with a Combined Approach

Product review mining is the process of extracting opinions of customers in reviews which are expressed by natural language. As the first phrase of product review mining, product feature extraction decides the quality of subsequent phrases. In this paper, we build a combined approach based on bootstrapping and ID3, ID3 is used as a feature selection algorithm in the iteration of bootstrapping. Given the seed set and classification feature set, the combined approach can automatically extract textual patterns with different structures, and avoid the design of textual pattern structures and the design of similarity function among textual patterns. We implement an automated product feature extraction system with the combined approach. Compare to previous study, our system achieves higher precision and better portability.

I. Introduction

Recent years, more and more customers release reviews on the web after using products. The reviews contain a wealth of information, which can help manufacturer to improve products and help potential customers to make a decision on whether to buy the product [1]. The number of customer reviews of a popular product can be hundreds and the reviews are mostly expressed by natural language. If anyone wants to get the information from customer reviews, he must look through the reviews one by one. As a rising field of information extraction, product review mining can acquire information from customer reviews automatically and quickly, so it can help people get rid of the hard job.