A Text Sentiment Classification Model Based On Multiple Multi-classifier Systems

With the rapid development of Internet technology,more and more people like to express their views on the Internet.Most of the texts published on the Internet contain certain sentiment tendencies and have potential commercial and social values,sentiment classification research for such online texts has also emerged.The research content of this paper is sentiment classification of texts.Aiming at this task,and inspired by a MCS(Multiple Classifier Systems)-based sentiment classification model,this paper proposes a text sentiment classification model,which is based on integrated multi-classifier systems.This model consists of three multi-classifier systems whose combination strategy is majority voting.Considering the defect of distinguishing feature selector in feature selection,a distinguishing feature selector based on class discrimination degree is proposed.The research content of this paper improves the classification accuracy of text sentiment classification from two aspects: optimizing classification model and optimizing feature selection method.The detailed research contents of this paper are summarized as follows:(1)A text sentiment classification model based on integrated multi-classifier systems is proposed.Unlike the traditional ensemble learning model,which uses single classifier as the individual classifier of ensemble model,the text sentiment classification model proposed in this paper uses three sub-multi-classifier systems as the individual classifier of the total model,and the three individual classifiers are heterogeneous.The first multi-classifier system uses SVM classifie as its individual classifier,which uses Bagging as the ensemble method of multiple individual classifiers.The second multi-classifier system also uses SVM classifier as its individual classifier,which uses RSM as the ensemble method of multiple individual classifiers.The third multi-classifier system uses NB classifier as the individual classifier,and using Boosting(AdaBoost)method as the ensemble method of multiple individual classifiers.The essence of the model proposed in this paper is to use ensemble learning to integrate multiple good and different individual classifiers to improve the classification ability of the model.(2)An improved feature selection method,distinguishing feature selector based on class discrimination degree,is proposed.As a global feature selection method,the distinguishing feature selector may filter out the features with strong ability of single class discrimination,a distinguishing feature selector based on class discriminationdegree is proposed,and the effectiveness of this feature selection method in many classification methods is verified by experiments.(3)The text sentiment classification model proposed in this paper is applied to the sentiment classification task of user reviews,and the sentiment classification process of user reviews is analyzed.On four datasets,different classification methods are combined with different feature selection methods.The experiments show that the proposed classification model and feature selection method are effective.