Feature Extraction and Opinion Summarization in Chinese Reviews

Abstract

The paper describes the process of mining opinions from Chinese reviews of products sold online. The struc- ture of Chinese reviews is free, which leads to a more complicated relationship between opinions and features. The paper introduces two main steps of opinion mining: feature extraction and opinion direction identification. The feature extrac- tion function first extracts hot features that a lot of people have express their opinions in their reviews, and then finds those infrequent ones. In order to improve the accuracy of the experiment, redundant features are removed. The opinion direction identification function takes the generated features and summarizes the opinions into two categories: positive and negative. We extract adjectives and negative adverbs as opinion words and use the Naive Bayes classifier to identify their direction. By direction, we mean whether an opinion is positive or negative.