Aspect-based Feature Recognition And Classification In Product Reviews

With the popularization of online shopping, there are more and more product reviews on the web, and product reviews mining is also going on.Product feature recognition is the basis of reviews mining, which provides evaluation objects for opinion analysis. The current methods of evaluation system construction are mainly defining the domain product evaluation system by manually, which are not universal. In this paper, we construct the network structure of feature words and create the evaluation system by extracting feature words and exploring the relations between different feature words through unsupervised method. The main contribution and research are listed below:1. The improvement of feature recognition by using ranking model.In this paper, the feature word recognition is divided into two parts:extraction and selection. The <feature, sentiment> words are extracted as the network nodes through the syntactic parser, then we use the PageRank algorithm to rank these nodes according to the relationship between the sentiment words and feature words, and select feature words with high scores, which improves the accuracy of feature recognition.2. Design method of how to construct product evaluation system.This paper constructs an aspect-based evaluation index by using the clustering method and a hierarchical structure evaluation system according to the membership degree and similarity degree between feature words. We decompose the relationship matrix by ISM method and constructs the product feature hierarchy structure. These two kinds of construction methods can be widely used in various product areas,providing a diversified evaluation objects for opinion mining.