Combining Semantic Comprehension and Machine Learning for Chinese Sentiment Classification

Semantic comprehension-based and machine learning based are two major methods for the classification of Chinese sentiment. The advantage of semantic comprehension-based method is that it can classify text among domains and achieve satisfied portability. However, the accuracy of classification is limited. Although the accuracy derived from supervised machine learning method is much better, the portability is rather poor due to randomly selection of samples and subjective labeling of semantic orientation. In this paper, a hybrid framework combining the advantages of the two methods was proposed. The text features were extracted preliminary based on semantic comprehension and were optimized by a novel information gain method. The features expressed in vector space model were integrated with traditional machine learning algorithm. Experiments show that support vector machine has the best discriminative power compared to other machine learning algorithms. Additionally, this framework improves portability and accuracy as compared to both semantic comprehension-based methods and machine learning based methods.