Research on Automatic Construction of Sentiment Lexicon Based on Bayesian Framework: Based on text sentiment classification

摘要

Most of the current sentiment lexicons only mark positive and negative polarities, and rarely considering the ambiguity of words that may lead to changes in sentimental polarity and intensity. This paper proposes a method on the construction of sentiment lexicon based on Bayesian formula. Considering the different parts of speech distribution factors, the sentimental polarity and intensity calculations of the words are converted into the positive and negative probability calculations of the under the Bayesian formula. Then, in order to synthesize prior sentiment knowledge and corpus sentiment knowledge based on Bayesian formula, this paper will construct a unified sentiment lexicon based on Bayesian framework and evaluate the lexicon through text sentiment classification tasks. The final experimental results show that compared to the existing sentiment lexicons, the effect of the sentiment lexicon constructed in this paper has been significantly improved, especially for the long text corpus, and the F1-measure reaches 77%.