Sentence Semantic Orientation Categorization Based On Semantic Role

With the rapid development of the Internet, the information provided by it increased sharply, especially in texts. In order to process this information effectively, there arise more and more missions for Natural Language Processing task. Semantic orientation categorization is one of these missions, which has been put forward recently and whose purpose is to adjudge the orientation of the commentary text for a given subject or viewpoint. Basing on the sentence level semantic orientation categorization, this paper will propose a new method to construct the sentence vector. Then we use Support Vector Machine to classify the semantic orientation.Firstly, we introduce the background, currant station and the theoretical basis of the new method mentioned above. Secondly, we introduce the Text Categorization, Point Mutual Information, Support Vector Machine classifier, semantic role and so on. They are all the basal work of semantic orientation categorization. Thirdly, we introduce the core work-six methods for semantic orientation categorization, four of which add the sentence construction information to the sentence vector by semantic role labeling. Lastly, we design interrelated experiment, get the result and analyze the problem and the defect of the system. The experimental data show that the method which takes the sentence construction information into account attains the accuracy of 88.97%. The experiment shows that this information can help to raise the performance of sentence semantic orientation.