摘要： 基于语料库的点互信息(PMI)计算方法依赖于语料库的完善性，基于HowNet的计算方法则依赖于知网相似度计算的准确性。为克服2种方法的局限性，提出一种HowNet和PMI相融合的词语极性计算方法，利用知网进行同义词扩展，降低情感词在语料库中出现频率低所带来的问题。实验结果表明，该方法的微平均和宏平均性能比传统方法提升约5%。

关键词: 情感分析, 点互信息, 知网, 同义词扩展, 相似度

Abstract: The polarity calculation of word level is the basis of sentiment analysis of sentence level and discourse level. The traditional calculation methods based on Point Mutual Information(PMI) or HowNet have their own defects: methods of PMI depend on the perfection of the corpus, and methods of HowNet depend on accuracy of the similarity calculation based on HowNet. In order to improve these deficiencies, an improved method for calculating the polarity of words is proposed, combining HowNet with PMI. First of all, HowNet is used to expand the synonyms of the emotional words in order to reduce the impact of some emotional words which have low frequency in the corpus, and then, according to the similarity calculation based on HowNet, it integrates the similarity based on HowNet with that of PMI. Experimental results show the new method increases micro average and macro average by 5% compared with traditional methods.

Key words: sentiment analysis, Point Mutual Information(PMI), HowNet, synonym expansion, similarity

中图分类号: 

• TP391