Incorporating an implicit and explicit similarity network for user-level sentiment classification of microblogging

Abstract

In Twitter, the sentiments of individual tweets are difficult to classify, but the overall opinion of a user can be determined by considering their related tweets and their social relations. It would be better to consider not only the textual information in the tweets, but also the relationships between the users. Previous approaches that incorporate network information into the classifier have mainly focussed on “a link” defined by the explicitly connected network, such as, follow, mention, or retweet. However, the presence of explicit link structures in some social networks is limited. In this paper, we propose a framework that takes into consideration the “implicit connections” between users. An implicit connection refers to the relations of users who share similar topics of interest, as extracted from their historical tweet corpus, which contains much data for analysis. The results of experiments show that our method is effective and improves the performance compared to the baselines.