Research On Sentiment Identification Of New Words In Weibo

With the rise of the concept of Web 2.0,the way of user generating content begin to develop rapidly,users can publish a lot of information on the Internet conveniently.In recent years,the applications of micro blog represented by Sina Weibo develop very rapidly.Because of the colloquial characteristics,Weibo has produced many new words.These new words in Weibo tend to have strong sentiment and express users'opinion,attitude and emotion on things.The research on sentiment identification of new words in Weibo has high social value,economic value and scientific research value.In this paper,we focus on the research of the method of identifying the sentiment of new words in Weibo.The method of based on context,the method of based on new word itself,and the method of based on rules are proposed and comprehensively applied.In the method of based on context,taking advantage of deep learning model on semantic learning,we modify the model according to the mission and use the model to learn the feature of context to identify the sentiment of new words.In the method of based on new word itself,we modify the word embedding model and propose the character embedding model in which the character is treated as a basic unit.This model effectively avoids the problem of insufficient learning caused by less appearance of new words.So it can achieve better word embedding and then improve the accuracy of sentiment identification of new words.In the method of based on rules,according to the characteristics of new words in Weibo,the method of recognizing entity names,the method of taking advantage of sentiment words in context and the method of processing homophonic words are proposed.Among them,the method of processing homophonic words analyzes the Chinese pronunciation and effectively improves the ability to identify the sentiment of homophonic words.In this paper,the experimental results of the three classifications of the sentiment of the new words in Weibo are given.By comparing with the results of other methods,the effectiveness of the proposed methods is proved.