Research And Implementation Of Multi-level Sentiment Analysis System

With the uprising of the Internet age, more and more people choose to express their views and opinions on things through the Internet. Sentiment analysis is a process during which the text containing emotional tendency is analyzed and automatically summarize emotional information contained therein. Sentiment analysis techniques plays an important role in monitoring public opinion, market research, trend forecasting and some other areas.The research herein stems from the need of a public opinion monitoring system, which utilizes sentiment analysis technology to cater for the judgment of public opinion monitoring. Sentiment analysis in Chinese now mostly focuses on the document-level sentiment analysis stage due to a late start; hence unnable to interpretate the emotion contained satisfiyingly. To solve these problems, this paper studies sentence-level sentiment analysis techniques based on ptimizing the traditional sentiment analysis methods. A method using dependency grammer is proposed to extract details of emotional expression. We design and implement a multi-level network sentiment analysis system, which not only provides fast real-time document-level sentiment analysis but also provides sentence-level sentiment analysis and give a good interpretation of emotion holders and emotional objects. The main contents include:1. Sentiment analysis methods based on text-classification. The performance of sentiment analysis based on different text-classification methods are given. These algorithms are optimized considering the specific features of sentiment analysis task and we propose a real-time coarse-grained sentiment analysis.2. We study the principle and implementaion of dependency grammer teniques, which can find Chinese syntactic structure capable of expressing emotion. By analyzing the impact of emotional words in the sentence and its context dependency of emotional expression, we propose a method of fine-grained sentiment analysis. The method is a sentence-level sentiment analysis method. While giving the emotional tendency in the sentence, it can also extract emotion holders and emotional objects.3. Based on the work above and web crawling technology, a multi-level network sentiment analysis system is designed and implemented. This system can give a document-level coarse-grained sentiment analysis and sentence-level fine-grained sentiment analysis with the ability to give emotional trend analysis after comprehensive analysis of the results of a large number of texts. Finally, the function, performance and speed of the system are tested.Simulation shows that the text is divided into positive, negative, neutral macro average F measure of 77.2% through the coarse-grained sentiment analysis system. During fine-grained sentiment analysis, the average sentence emotional tendencies macro F metric analysis is 70.4%.The recognition rates of emotional holders and objects are 81.6 % and 72.2 %, respectively. The system can satisfy the needs of sentiment analysis of different levels.