Sentiment Analysis Of Weibo Texts Combined With Emoticons

Weibo,as a comprehensive social platform,is sought after by the general public due to its diversity and effectiveness,it becomes one of the most popular social applications at present.Weibo contains a large number of subjective texts with emotions.The sentiment analysis of Weibo data can be used to understand the public attitudes toward social events and users' comments on products,which are of great significance to product research and the monitoring and early warning of public sensation.Weibo text and emoticons affect the overall emotional orientation of Weibo.However,at present,most researches on the sentiment analysis of Weibo do not attach importance to emoticons.Emoticons not only have their own emotional orientation,but also affect the emotional strength,even emotional orientation of Weibo.Therefore,in order to judge the emotional orientation of Weibo more objectively and accurately,this paper combines the characteristics of the emoticons in Weibo to conduct sentiment analysis of Weibo.The research mainly includes the following three aspects:(1)Constructing Weibo corpora: To conduct sentiment analysis of Weibo,a large number of well-targeted Weibo corpora and a large number of manual text annotations are required.At present,there are no Weibo corpora with lots of emoticons and annotations that are suitable for this study.Therefore,this paper constructs a Weibo corpora in a targeted manner.And this corpus contains a large number of emoticons.At the same time,an automatic tagging method using emoticons and emotional words is proposed to reduce the workload of manual tagging.(2)Researching emoticon vectorization algorithm: We need to extract features from emoticons because emoticons are one of the most important considerations in the analysis of Weibo's sentiment.Therefore,this paper presents a emoticon vectorization algorithm.First,we define the emoticon vector and description vector,and model the similarity between the two vectors.Then we use gradient descent algorithm to determine the vector form of the emoticon.This algorithm can retain the semantic information of the emoticon.Meanwhile,this algorithm makes the emoticon as well as vocabularies,which can be applied to all fields of Natural Language Processing.(3)Sentiment analysis of Weibo based on emoticons: The network structure of the convolution neural network determines that it can extract the deep semantic information of Weibo texts and emoticons.In this paper,we use the classification model of convolutional neural network and uses emoticon vectors and word vectors,sentiment classifiers are trained and then performing Weibo sentiment analysis.The results show that compared with the pure text sentiment analysis that ignores the emoticons,the Weibo sentiment analysis that integrates the emoticon information into the Weibo text can improve the accuracy of the Weibo sentiment classification.