Research And Implementation Of Sentiment Search Engine Based On Topic Model

With the popularization of web and the rapidly development of web2.0, the web has become the largest common data source in the world. At the same time, the dramatically development has brought great convenience to people. More and more people prefer to choose shopping online. Usually, in such trade procedure, the customer can only evaluate the quality comprehensively by the reference diagrams which the storekeeper provided and reviews of the other people. So, the online review has become the most important evaluation. The task of opinion mining is extracting the opinion from reviews and analyzing the sentiment orientation. Users always publish their reviews according to some special attribute of the topic. For a commodity, there are several attributes. At the same time, lots of feature words can be used to describe an attribute. People usually use different sentiment words according to the different attribute. As the starting point of the relationship of topic attributes and feature words and sentiment words, in this thesis, a topic model was proposed. It was applied to achieve the sentiment analysis. With the current requirement, in this thesis, a sentiment search engine based on topic model is studied and implemented.In this thesis, the concept of topic model is introduced and an algorithm is proposed to build the topic model. Firstly, feature words and sentiment words are extracted, respectively by language mode rule and sentiment dictionary. Secondly, the feature-sentiment-word-matrix is initialized according to feature-sentiment-word-pairs. Then the matrix is transformed to the attribute-sentiment-matrix by feature word similarity. At last, the topic model is built.Meanwhile in this thesis, a framework is designed to analyze sentiment orientation based on topic model. Firstly, the reviews are processed as documents. Then the reviews are classified to different attributes by topic model. At last the sentiment scores of reviews are evaluated by opinion dictionary.Finally a sentiment search engine IMovie based on topic model is designed and implemented.The experiments show that the algorithm can perfectly build topic model, and the sentiment analysis framework has a very good precision, recall and F-measure. At last, the sentiment search engine called IMovie can satisfy the most of requirements of people.