Text Sentiment Analysis Based On Statistical Knowledge

With the rise of the Internet and the advent of Web3.0era, web texts contain more and more personal reviews. Unfortunately, Mainstream news coverage is the mixture of junk information and valuable comment information, especially for the comment about some events and products. Therefore, it’s quite difficult to obtain the valuable reviews about those events and products from the countless web texts, and that is why text emotional tendencies research appeared.The main task of text emotional tendencies research is identifying subjective contents or objective contents from a semi-structured or unstructured text, and determining the tendency of these texts, and then exploring the expression of objects, role objects and emotional expression contents. Text emotional tendencies research can be conducted by two basic strategies. One is based on the characteristics of syntax and semantic rules, and the other is based on the statistical knowledge-based text mining work.This thesis introduces statistical machine learning strategy into text emotional tendencies research. Different from the general strategy of emotion research work, this strategy is based on the evaluation object to identify the emotion content. In the effective expansion of the word, this paper present a feasible based on adjacent word derived object recognition strategy. In the mining of the evaluation content, the thesis applies a semi-supervised statistical learning algorithm. This algorithm merge the Chinese basic semantic pattern, the pattern’s information gain as well as the context word distribution characteristics together. What’s more, the whole research work uses three fields corpus for cross experiment.