Emotional sentence identification in a story

Abstract

In this paper, we investigate the methods of fusing different classifiers to identify emotional sentences in text. The Extreme Learning Machine (ELM) and Support Vector Machines (SVM) are two classifiers used to predict a sentence neutral or emotional. We use the UniGram, subjective words, and special punctuations, etc. as features. A method of calculating emotion value of a word is presented, and the values are employed to compose the features of an emotional sentence. To further enhance the system performance, we divide the features into three subsets, and train different models of the two classifiers on each feature set. The six models are then combined through a weighted summation fusion method and FoCal fusion method. We evaluate the system performance on a corpus of children's tales, and the experimental results demonstrate that the fusion of models can improve system performance. © 2012 ACM.