Abstract:

The performance of sentiment analysis can be improved effectively with the help of a wide-coverage and good domain-adapting sentiment lexicon. We firstly design two Chinese sentiment lexicon extension algorithms, which  base on conjunctions feature and POS-vector statistical feature respectively. We then propose an integrated mixing feature method that combines the two algorithms. Fine-grained positive and negative values can be calculated for opinion words, the coverage of the lexicon can be improved within a domain, and the adaption of the lexicon can be improved with adjustment in the domain. Experimental results show that the extension lexicon has wider coverage and better adaption than a general lexicon in a domain, and the proposal's performance of sentiment classification can approximate that of a supervised method.