Research On The Method Of Sentiment Analysis In Weibo

Weibo as the current information dissemination of the carrier,so that people can more easily comment on hot events,express their feelings.In particular,after the introduction of sina weibo,the number of Chinese weibo users increased,the amount of data also showed an exponential growth in weibo.Weibo text sentiment analysis is conducive to public opinion monitoring,marketing and rumor control.At present,the weibo text sentiment analysis mainly concentrates on the polarity judgment,namely positive,negative and neutral.This is far from the purpose of opinion mining.Therefore,it is necessary to study the fine grained sentiment analysis method for short text of weibo.This article divides weibo emotion into four categories: happy?sad?angry and disgust.This paper studies the following three aspects:(1)In text classification,CHI is a good feature selection method.This paper analyzed the shortcomings of the CHI feature selection method,and proposed a CHI feature selection algorithm based on intra class and inter class distribution factors.By introducing the above two factors,they can reduce the interference caused by the low frequency words,and the negative contribution to the classification when the feature words are evenly distributed in each category.In addition,the traditional CHI method is in the global scope of the feature selection,ignoring the correlation between the feature and the category,so the feature selection method based on category was proposed.(2)Considering the influence of feature selection on the feature weighting,the improved CHI method in the step(1)was combined with the commonly used feature weighting algorithm TF-IDF to form a new feature weighting algorithm.Experimental results show that the new feature weighting algorithm can effectively improve the accuracy,recall and F measure of weibo sentiment classification.(3)This paper studied the current mainstream ensemble learning methods,and compared the performance of the five ensemble learning methods: Bagging?Boosting?Stacking?Random subspace and Random forest,which were used in the field of weibo sentiment analysis.At the same time,for the best performance of the Random subspace method,the classification performance was compared when using different base classifiers.