Maximum entropy based emotion classification of Chinese blog sentences

Abstract

At present there are increasing studies on the classification of textual emotions. Especially with the rapid developments of Internet technology, classifying blog emotions has become a new research field. In this paper, we classified the sentence emotion using the machine learning method based on the maximum entropy model and the Chinese emotion corpus (Ren-CECps)*. Ren-CECps contains eight basic emotion categories (expect, joy, love, surprise, anxiety, sorrow, hate and anger), which presents us with the opportunity to systematically analyze the complex human emotions. Three features (keywords, POS and intensity) were considered for sentence emotion classification, and three aspect experiments have been carried out: 1) classification of any two emotions, 2) classification of eight emotions, and 3) classification of positive and negative emotions. The highest classification accuracies of the three aspect experiments were 90.62%, 35.66% and 73.96%, respectively.