Multi-source Emotion Tagging for Online News

With the rapid growth of social media and online news services, users nowadays can respond to online news by rating subjective emotions such as happiness, surprise or anger actively. Once the user ratings is over a certain range, it begins to show up a tendency of what most people think and feel, which can help us understand the preferences and perspectives of most users, and help news providers to provide users with more positive news. Thus it has become a pregnant research problem to tag emotion automatically. This paper tackles the task of emotion tagging for online news with multi-source including news article and comment, as emotion is not only tagged after reading news article, but also can be incorporated in comment with what they feel. In this paper, a novel classification model are proposed with two layer logistic regression. The new approach get outputs from basic classifiers and combine them in a new classifier, making a more accurate prediction when compared with a single source method. An extensive set of experimental results on a real dataset from a popular online news service demonstrate the effectiveness of the proposed approach.