An approach to sentiment analysis of short Chinese texts based on SVMs

This paper uses a machine-learning method to determine the sentiment polarity of short Chinese texts. Firstly, a new way to extend the sentiment dictionary is presented. The sentiment dictionaries from NTU and HowNet are extended by using the word2vec tool provided by Google. The review texts are collected from Internet as datasets. Then the feature weight of the words is enhanced, including the words that appear in the sentiment dictionary that has been extended and the words next to the sentiment words. The reviews are classified into two classes, the positive semantic orientation and the negative semantic orientation. The result of experiment shows the progress in the accuracy.