Chinese Sentiment Orientation Analysis

• Published in: 2010 International Conference on Computational Intelligence and Security pp. 1 - 5
• Main Authors: Wenbin Pan, Yanquan Zhou
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2010
• Subjects: Algorithm design and analysis; Classification algorithms; Dictionaries; Learning; negative; neutral; polarity; positive; sentiment orientation; Support vector machine classification; Thumb; Training
• ISBN: 9781424491148; 1424491142
• DOI: 10.1109/CIS.2010.8

In this paper, we present one new method to analyze and classify the sentiment orientation of merchandise comments into three categories: neutral, positive and negative. Nowadays, many methods can be used to achieve this goal, however, we find that those methods may work well in dividing the polarity sentences into positive and negative but may not have a good result on neutral sentences, so a divide and conquer strategy is applied to firstly classify the texts into two parts as neutral and polarity texts. Then the polarity texts are divided into positive part and negative part. In the first step, TSVM tool is used to achieve the neutrality and polarity classification, but the training data used in our work is very special, which contains many polarity sentences but very few neutral sentences, so the strategy is adopted to divide the polarity data into several small parts, and each part polarity data is combined with all neutral data as training data, by this way several TSVM classifiers can be obtained, and by voting scheme the final result can be gained. In the second step, we propose an algorithm to achieve positive and negative classification. Firstly, a method is designed to re-evaluate each sentiment word and divide the dictionary into two parts based on confidence, which can reduce the negative impact of low-confidence words. Then an orientation analysis and classification algorithm is proposed to classify the polarity sentences step by step. Meanwhile, a set of rules is also built to classify those sentences which contain sentiment words that appear not in our sentiment dictionary.