Sentiment Analysis Of Chinese Weibo Texts Enhanced With Emojis

With the development of network and information technology,Weibo has gradually become the most popular social platform due to its effectiveness and freedom.There are various kinds of topics on Weibo,and users can share their opinions and attitudes conveniently and quickly.Therefore,massive amounts of subjective data have accumulated on Weibo.Through the sentiment analysis of these Weibo data,we can mine the public's emotional attitudes to hot topics,or likes and dislikes of a certain commodity,which is of great value for many areas such as production,advertising,and public opinion supervision.At present,most of the researches on Weibo sentiment analysis only focus on how to deal with the text and use the text to mine the emotion.However,a large number of comments on Weibo contain a variety of emojis,which together with the text represent the emotions expressed by users.In order to improve the accurate of emotional analysis for microblog data,this paper conducts the following research works based on the characteristics of Weibo comments:(1)We proposed an emoji vectorization algorithm.In the current popular sentiment analysis methods,how to embedding emoji as a feature vector that the model can recognize is very important.This article first builds an emoji network using Weibo comments,the purpose is to obtain the emotional connection between emoji.At the same time,emojis were portrayed from multiple emotional angles.Then use the manually marked initial emotion and the emoji network to obtain the final emotion value of the emoji.Finally,the corresponding emotion word description is obtained through the emotion value of emoji,and then the word embedding method is used to embed the expression characterization words to obtain the final emoji vectors.At the same time,a microblog corpus has also been constructed.Due to the late start of research works on Chinese microblog text sentiment analysis and the neglect of emojis by most research works,it is currently difficult to find a microblog corpus suitable for our research.In response to this situation,this article collected a large number of micro-blog comments containing emoji expressions,and at the same time some data of it were manually annotated to provide data support for subsequent research work.Finally,the experiment verifies that the emotion information refracted by the emoticons is retained to a certain extent,and the emoticons with similar emotion expressions are mapped to similar positions in the vector space.This laid the foundation for embedding emoji information into the sentiment analysis model based on deep learning..(2)A neural network sentiment classification model enhanced with emojis was proposed.In recent years,a lot of research work has proved the superiority of deep learning technology in sentment analysis.Therefore,this paper proposes a neural network emotion analysis model,uses BiLSTM to extract the text representation of comments,and proposes a new attention mechanism to obtain the auxiliary representation based on emojis.Finally,the text-based representation and emoji-based auxiliary representation are merged into the final sentence representation vector,which is sent to the emotion classifier for sentiment analysis of Weibo comments.Through a series of experiments,the performance of the emotion analysis model proposed in this paper is proved to have better performance.