Reader emotion prediction using concept and concept sequence features in news headlines

Abstract

This paper presents a method to predicate news reader emotions. News headlines supply core information of articles, thus they can serve as key information for reader emotion predication. However, headlines are always short which leads to obvious data sparseness if only lexical forms are used. To address this problem, words in their lexical forms in a headline are transferred to their concepts and concept sequence features of words in headlines based on a semantic knowledge base, namely HowNet for Chinese. These features are expected to represent the major elements which can evoke reader's emotional reactions. These transferred concepts are used with lexical features in headlines for predicating the reader's emotion. Evaluations on dataset of Sina Social News with user emotion votes show that the proposed approach which do not use any news content, achieves a comparable performance to Bag-Of-Word model using both the headlines and the news contents, making our method more efficient in reader emotion prediction. © 2014 Springer-Verlag Berlin Heidelberg.