Research On Sentiment Classification Method Of Chinese Reviews

Review texts on the websites often contain positive or negative emotions,support or opposite attitudes.The analysis of those emotions and attitudes in review texts has significant social value and commercial value.Although the research on the method of sentiment classification for Chinese reviews has made some achievements,it still needs to be improved and perfected in the subjective and objective text classification method and the subjective text sentiment classification method.In research of the subjective and objective text classification method,this paper puts forward the standard for tagging the subjective text of the reviews.Features based on N-Gram are selected as semantic features and an N-POS feature is used as syntactic feature.Considering that TF-IDF meets the requirements of the feature selection method,TF-IDF is studied as a feature selection method,as well as chi-square statistics and information gain.Three classic classification algorithms are used.On this basis,a method based on the similarity of classification results is proposed for the selection of feature which is combined into joint features.Based on this feature combination method,the subjective and objective text classification is completed,and the objective text which may exist in the reviews is eliminated,which lays the foundation for the sentiment classification of the follow-up subjective text.In research of the subjective text sentiment classification method,a feature based on POS and features based on N-Gram are selected,then TF-IDF,chi-square statistics and information gain are studied as feature selection method.Considering that classification algorithm is domain-dependent,this paper seeks to integrate different types of classification algorithms to overcome their individual drawback to complete the task of the subjective text sentiment classification.During this process,diversity of the ensemble learning is studied,then an ensemble strategy is proposed for multi-classifier,finally sentiment classification based on ensemble learning for subjective text is completed.A wide range of experiments are conducted for proposed methods,and suitable evaluation index for the methods is designed.The result of experiments shows the effective of the subjective and objective text classification method and shows that sentiment classification method based on ensemble learning for subjective text improves the classification performance.