Research On Twitter Text Sentiment Analysis Based On Progressive Transductive SVM

In recent years, with the rapid development of Internet technology and the further popularization of mobile terminals, Social Network Service (SNS) has penetrated people’s daily lives gradually. Twitter is a typical social network and micro-blog service site and it is one of the top ten visits sites of the world. Users can update their status at any time and follow others at the same time. Hundreds of millions of tweets everyday record whatever the users seen, heard, done and felt. And they share their joy, anger, sadness, and joy in this way. Mining sentiment expressed in tweets has an important role in marketing, public opinion monitoring, emergency response and other aspects.Affective computing has become a hot research field of artificial intelligence. Text sentiment analysis is different from traditional text classification because sentiment is subjective, obscure and there is no unified criteria. And the length limit, colloquial language style and full of noise make Twitter sentiment analysis more challenging. Several essential issues are researched including tweets preprocessing, feature selection, semi-supervised sentiments analysis algorithm and so on.A series of modified text preprocessing methods are researched to reduce the effect of noise in tweets and the work focuses on preprocessing on URLs, negation, repeated letters and so forth. And the role of preprocessing for Twitter sentiment analysis is verified by comparing experiment. The ability of feature selecting of different standards is also compared especially for DF, IG and χ2 Statistics. The experiment results illustrate that preprocessing and feature selecting can improve the performance and reduce the feature space dimension at the same time. In addition, a novel semi-supervised Twitter sentiment analysis algorithm based on progressive transductive SVM is researched in order to overcome the difficulties of obtaining massive annotated data and avoid wasting of unlabeled data. The algorithm improves the performance of sentiment analysis by introducing the interference factor. The algorithm is also adaptive for data distribution and the progress of study and training time are automatic control.