Research On Social Text Sentiment Analysis Based On Weibo Platform

With the development of information era,social network has been one of the major communication ways in people’s daily life.Among various social platforms,weibo is favored by young people due to its flexibility and openness.Due to the huge value of short texts generated in social networks,sentiment analysis for short text has attracted more and more attention from academic and industry.However,with the emergence of emoticons,people prefer to use emoticons to express their emotions.Since the emoji are added to the new version of i Phone by Apple Inc in 2012,it sweeps multiple social networks(such as weibo and twitter).From then on,the majority of weibos are consist of texts and emojis.Therefore,current sentiment analysis methods which only consider pure texts are not suitable for texts containing emojis as the result of semantic information loss.Based on previous deep learning model,this dissertation takes the feelings contained in emojis into consider,rather than only regards emojis as the auxiliary features for sentiment analysis.This is an effective way to mine emotional tendency of emojis and increase the performance of social texts sentiment analysis.The main works are summarized as follow:(1)For the problem of lacking emojis in current commonly used datasets,this dissertation constructs a real world dataset containing emojis,named Weibo＿SL.This paper first uses data preprocessing technology to clean the text data of social networks,and uses manual methods to filter data and annotate emotions.Finally,relevant experimental analysis is conducted to prove the validity of this data set.This dataset contains 6436 weibos from 100 different users from the January of 2017 to the July of 2020.Among these weibos,5526 of them contain both texts and emojis,and they can reflect the real environment of social network texts.(2)For the problem of current sentiment analysis models consider less on social network texts containing emojis,this dissertation proposed an sentiment analysis method based on three level attention mechanism.Aiming at the text data of social network containing emoticons,the scheme deeply mines the emoticons’ emoticons and emoticons’ emoticons,and analyzes the overall emotional state of the text based on the emoticons’ emoticons.This method proposed a novel weighting method for word embedding based on TF-IDF and Chinese emotion dictionary,which can significantly increase the attention of emotion words.In addition,this method utilizes Bi-LSTM and attention mechanism to effectively mine emotion semantics in texts and emojis,and further analyzes the emotion tendency of whole sentence.