Research On Sentiment Classification Of Real-time Comments Based On Sentiment Tendency Clustering

This work is partially supported by the National Natural Science Foundation of China(Project number:71673032).The rapid development of social networks has greatly influenced the Internet for generating data,People share their opinions or experiences through the Internet and express various of sentiment orientations such as favor,anger,sadness,happiness,criticism,praise,etc.This commentary information is rapidly generated and accumulated,providing valuable research data for sentiment analysis.This type of commentary data is quite complex due to the arbitrariness of human language expression,and as time goes by,large number of network terms and new vocabulary are appearing on the network.Meanwhile,the polysemy and semantic change also caused great difficulties for sentiment analysis.The current application of sentiment analysis can only be carried out from semantic and grammar,and it's not effective enough.Therefore,it is necessary to start a new perspective on sentiment analysis.In the previous work of this paper,some special user comments with high real-time performance were found,the evaluation objects of these comments will change over time and have the same sentiment orientation in the same time period.The concept of real-time comment is proposed,in order to distinguish it from traditional user comments and make use of this real-time performance in sentiment analysis.Further,the two features of real-time comment were found by counting and mining of large amounts of data:(1)The generation of comments is intermittent;(2)Comments in one period have the same sentiment orientation.The features are greatly inspiring the follow-up research.In order to apply the clustering and classification for the sentiment orientation of real-time comments,the Sentiment Orientation Clustering Algorithm based on Quantum Harmonic Oscillator Model(SOCA-QHOM)and the Modified Bayesian Algorithm based on Sentiment Orientation Clustering(MBA-SOC)are proposed.SOCA-QHOM projected the real-time comments into time axis,and the time axis was divided into intervals to form a discrete objective function,then the clustering was transformed into function optimization.Clustering was carried out by using the probability interpretation of the transition of a harmonic oscillator from high energy to low energy.MBA-SOC provide an improved Bayesian model by using the sentiment orientation of real-time comments and replace the priori probability of Bayesian model.The improved model corrects the Bayes model when processing sentiment classification.Finally,comparative experiments were deployed to validate the effectiveness of the algorithms.Experimental results show that both SOCA-QHOM and MBA-SOC are effective.