Research On Computation Of Chinese Text Sentiment Classification Of Multi-granularity

Text sentiment classification and discrimination through the automatic division of the comments positive or negative comments to tap customer feedback for a product, and judgment the classification of them. Text classification of objects related to sentiment words, sentences, paragraphs and chapters. The complexity of the Chinese language tend to make words, sentences and text of the sentiment detection studies there have been many difficulties, Because in the text field of affective computing category, usually granular levels of various research methods are different, for some understanding of linguistics is not enough detailed, resulting in discrimination of certain sentiment and subjective judgments appear inconsistent situation. Therefore, this article on the basis of first in HowNet, improve the method of word sentiment, then according to linguistic knowledge of the various factors that defines the sentiment quantitative. Finally, the sentence topics extraction and polarity discrimination method to improve, and design with rule-matching algorithm applied to the calculation of the sentence to the sentiment orientation.This paper mainly include follow aspects:(1) The hownet theoretical premise, based on the Chinese tend to the sentiment Words calculation methods improved. Specifically, eliminating the concept of the hownet in the original description of dislocation and the emergence of the concept of ambiguity, This part of the work was conducive to sentiment orientation to judge a sentence.(2) Presented pairs of sentences tend to study sentiment considerations and quantitative analysis. Will be extended negative words, and words with semantics orientation tend to the details of the modified polar quantitative, further study appears in the negative sharing sentences in Chinese, as well as the negative polarity of the transfer of more sentiment issue of sentence, because these factors also tend to affect the degree of re-sentence the sentiment orientation; Then, detailed analysis of the exclamation, respectively, with the rules and interjection, start from the linguistic point of view, on the possible form and exclamatory expressions of various sentiment to analysis the sentiment.(3) Improved the polarity of a sentence determine the specific method, No longer rely solely on the dependency structure of the relationship between syntactic structure to transfer of polar values, but according to the definition of the relationship between dependency structures extracted sentence topic, Dependency grammar comprehensive consideration the distance and the impact of polar modifiers, while the rules of the sentence to a negative match, making the sentence level, the polarity of the discriminant analysis more accurate. Next, in the words, sentences tend to study the basis of sentiment, sentiment classification for Chinese text specific system design and application.Finally, for this paper, the word, sentence sentiment tend to the methods used the system of measurement, comparison with other research methods based on HowNet, experimental results show that the method proposed in this paper the validity, accuracy and precision are improved.