Text Orientation Analysis On Sentences And Documents

The purpose of text orientation analysis is to determine the attitude or views expressed in text, in recent years orientation analysis is becoming the hot issue of information retrieval and natural language processing. Orientation of the text is divided into two areas: emotion, and sentiment (opinion). Both are a reflection of the Subjective will of people, emotion express people's own emotional ups and downs, such as happiness, sadness, etc; sentiment express the attitude and degree of people to the outside world or things. Such as support, opposition, etc.Text orientation analysis widely used, in public opinion analysis, harmful information filtering, film evaluation, investigation on products has wide application prospect, but the public corpus of text orientation is still small, that is the bottleneck of research. Analysis of text orientation is quite preliminary, to make sentiment analysis to the extent applicable, there is still a long way to go. In this paper, I am focused on the emotion analysis of sentence and sentiment analysis of document.Firstly, this paper presents a method to analyze sentiments on Chinese sentences, where the sentiments include happy, angry, sad, and fear. word features in sentences were extracted on the basic of the sentiment lexicon and the sentiment polarity lexicon, and word sequence features were extracted by rules while processing sentiment analysis on sentences, then ME model was used to classify the sentences. Good performance of sentiment analysis was gained in COAE 2009.Secondly, sentiment classification is the main method to analyze sentiment polarities on documents, but one of deficiencies is that it is difficult to integrate the structure features. A cascade model based sentiment polarity analysis is proposed to overcome the deficiency. The model consists of two levels: the clause level and the document level. The document is broken orderly into clauses which are classified into positive and negative categories by Maximum Entropy model, these categories are then combined with types and positions of clauses as features for document classification with Support Vector Machine model. Meanwhile, a Single-label Cascade Model based on cross-validation is proposed to avoid tagging on the clause level. Compared with traditional methods of sentiment classification, the accuracy is improved by 2.53%.Finally, in this paper, the experiment of the sentence emotion analysis and document sentiment analysis are analyzed.