Chinese Opinion Sentence Extraction Based On SVM Classification

With the increasing amount of internet information, large redundant information makes it difficult for us to obtain valuable information accurately and quickly. Especially in the large-scale online shopping boom, the majority of consumers will look up other users'evaluation online for guidance before shopping. The extraction of view sentence is to distinguish the subjective opinion sentence and objective opinion sentence, and then the consumers could search for the evaluation of the products quickly, meanwhile the manufacturers could carry on the market research of the products. Based on the above situation, extraction of Chinese opinion sentence becomes a research focus in the processing of Internet intelligent information. It includes natural language processing, information retrieval, information extraction, machine learning and other contents in multiple fields.In this paper, the main research contends contain the followings: firstly, through summarizing the language features of opinionated sentences, three syntactic structure templates on the basis of indicative verbs are proposed while the confidence of the templates are proved by experiments; then a kind of dependent relationship extraction algorithm is designed to extract the dependency relationship between the head word and the affiliated word of head word in Chinese opinionated sentences and to dig out the dependency relationship templates. Thirdly, by using SVM classifier and selecting some features based on words and POS for sentence binary classification, and by choosing different feature dimensions to find the most optimum and effective features combination. Finally, by cascading the syntactic structure templates, the dependency relationship templates and SVM classifier to construct the opinionated sentences extraction system, which ensures the most effective result.The original points in our paper are listed as followings:1) the effective dependency relationship extraction algorithm is proposed; 2) the syntactic structure templates, the dependency relationship templates and SVM classifier are cascaded, and by matching the templates can complements the incorrect classification resulted from the probability and statistics machine learning model to make mutually complementary.