Abstract:

To solve the problem of Chinese microblog sentiment classification, a sentiment classification method combining PMI and Hownet based on microblog data is proposed. Through the research on the short and novel features of microblog data, a method of dictionary merging is proposed.The existing dictionaries are merged according to Hownet word similarity, and PMI is used to perform sentiment classification of online words. The network sentiment words are added to construct sentiment dictionary that adapts to the features of microblog text, and sentiment classification models are trained based on the new dictionary combined with supervised learning methods. The experimental results show that using this method for sentiment analysis can effectively identify the impact of new internet words on sentiment analysis, with an accuracy rate of 78.3%. In the sentiment analysis of microblog containing new words on the Internet, the accuracy rate is higher than that of only using dictionaries or supervised learning.

Key words: sentiment dictionary, microblog text classification, supervised learning, sentiment analysis, Hownet similarity, PMI, opinion mining, benchmark words