Multi-label text classification approach for sentence level news emotion analysis

Abstract

Multiple emotions are often evoked in readers in response to text stimuli like news article. In this paper, we present a novel method for classifying news sentences into multiple emotion categories using Multi-Label K Nearest Neighbor classification technique. The emotion data consists of 1305 news sentences and the emotion classes considered are disgust, fear, happiness and sadness. Words and polarity of subject, verb and object of the sentences and semantic frames have been used as features. Experiments have been performed on feature comparison and feature selection. © 2009 Springer-Verlag Berlin Heidelberg.