Sentiment Analysis of Chinese Micro-Blog Based on Classification and Rich Features

With the development of the Web2.0, micro-blogs gradually become a common essential part of the public life. The reviews in the micro-blogs have huge hidden value. Many machine learning approaches have been used to solve sentiment analysis. However, the features used in existing researches are still not enough. To improve the accuracy of sentiment analysis, in this paper, we use a classification approach to solve two tasks of sentiment analysis: identifying opinion sentence and judging sentiment polarity of the emotional sentence. And we incorporate five kinds of features: sentiment lexicons-based features, N-POS(part of speech combination)-based features, pattern-based features, special symbols-based features and length-based features to train seven classifiers and compare their performance. Experimental result shows that Random Forest classifier achieves the best performance.