Research on the Model of Multiple Levels for Determining Sentiment of Text

Sentiment analyzing has been used in many fields, such as information security and evaluating products on web. In this paper, we propose a new model of multiple levels using semantics analyzing and the conditional random fields techniques to determine sentiment of a text. Sentiment of a document is divided into two parts in this model: global sentiment which is the sentiment of the entire text and local sentiment which is the sentiment associated with a particular part of the text. All information of local sentiment determine the global sentiment of text. According to this new model, a text is separated to several semantic paragraphs based on semantic similarity, and sentiment of semantic paragraphs is defined as local sentiment. Global sentiment of the text is identified by analyzing local sentiment information. Experiments results demonstrate that the performance of this fine granularity model is better than that of traditional SVM method.