Research On Phrase Pattern Based Sentiment Extraction For Reviews

Sentiment classification is a branch of natural language processing, which use sufficient semantic information of text. It is valuable in the fields of text information retrieval, survey on network and supervision by public opinion and becomes hotspot of intelligent information processing.Most of the methods of text classification are topic-based classification based on key words. That can't satisfy people's personal requirements on text Orientation. To deal with it, a semantic oriented sentiment classification approach was proposed based on pioneer researches. This approach classifies text by constructing phrase patterns with the words which can express the sentiment orientation of the author and analyzes complicated sentence in semantic layer. The main research work and achievements are:1. We extend four basic emotion (happiness, anger, sorrows, joyful) using six types emotion (happiness, anger, disgust, fear, sadness, surprise) in psychics' domain. Aiming at characteristic of reviews, we classify emotion into four fine grained types, which are praise, despair, anger and hate.2. We propose a semantic oriented sentiment classification approach to classify the text. We consider complex sentence and extract phrase patterns which are up to syntax structure. Then we integrate semantic orientation Latent Semantic Analysis and Semantic Orientation (SO) mining method based on PMI-IR algorithm to compute sentiment value of words and phrase separately. The result testifies that this approach improves the inadequate of distinguishing sentiment orientation of the past methods because of lack of semantic constraints. The phrase pattern based sentiment classification has achieved the precision ratio of 95% and the recall ratio of 93%.3. We import emotion space theory in analyzing sentiment orientation of text. It can help us effectively describe fuzzy emotion by quantificational mapping the fuzzy emotion into the emotion space.4. We construct a dictionary of the special words and the pet phrase by dynamic study to increase the precision ratio of our method from 89% to 95%.