Selecting clause emotion for sentence emotion recognition

Sentence emotion recognition allows for deeper analysis of textual emotion. Based on the finding that sentence emotional focus can be expressed by some clauses in this sentence, this work proposes to select clause emotion for sentence emotion recognition. In the first step, a Maximum entropy (MaxEnt) classification model has been built for word emotion recognition. In the second step, a homogeneous Markov model (HMM) classification method is used for clause emotion classification. In the third step, nine text features are selected, and genetic algorithm (GA) is used to specify the weight of each text feature. The sentence emotion is an addition of all selected clause states in this sentence. The experimental results showed that there are 9.1% and 3.6% improvement for the two tasks respectively when comparing with the baseline. It is demonstrated that clause selection is able to improve the performance of sentence emotion recognition significantly.