Sentiment Analysis Based On Sentence’s Structural Features

Nowadays,with the development of the data mining and machine learning, the field of sentiment analysis are caused much public concern. Sentiment analysis which has great commercial value and research value can be used in many fields including public opinion analysis, product reviews analysis and so on. The main research object of the paper is the tendency judgement of the sentence level. Given a sentence, we should determine whether it is positive or negative tendency. The main contents of this paper include:(1)At the beginning, the research uses an investigation to analyze the present situation of sentiment tendency analysis, focusing on the methods based on syntactic rules and machine learning, then spots the advantages and disadvantages of these two methods. Basing on the analyze, the research framework of this project has been put forward, that is, to achieve the baseline system based on syntactic rules and to propose an quantitative method to prove the sentiment method and the applications of the method.(2)Based on the calculation of syntactic rules, a sentiment analysis system for product attributes was designed and achieved. The whole system consists of the following four modules:real-time pre-processing module, the dictionary generation module, product attributes clustering module, sentiment intensity calculation module. The system can effectively analyze the data of the English comment of the Amazon online shop.(3)According to the disadvantages of syntactic rules algorithm, a new model named structured emotional vector space model(SEVSM) is found. This method can simplify the syntactic dependency tree to sentiment lable triples and present all the triples to sentiment feature vectors. The advantages of this approach is that both can make full use of syntactic rules, and can be structured unstructured emotional information, and help to directly use machine learning algorithms.(4)In order to verify the feasibility and effectiveness of SEVSM, we used it in all kinds of experiments. The sentiment vector is directly used in many machine learning algorithm which performs well. The results of the experiments on the sentiment tag optimization and features optimization have been compared. Experiment results shown that our proposed new model is able to achieve a rather performance cooperating with various classic machine learning algorithms directly, and it has much potential to improve the accuracy of sentiment analysis results.