Abstract
Due to the difficulties of building emotion model for a dialogue agent directly from a dialogue sentence, we propose a two stage by mapping words in a dialogue text sentence into thematic their roles and then inferring the embedded emotion from the dialogue sentence based on the thematic information. Therefore, in this thesis, we focus on the learning classification problems: one is the thematic role assignment based on syntactic structure information and features of words of a sentence and the other the emotion detection based on thematic role information assigned to a sentence. We integrate AdaBoost and neural networks learning approaches to train the two classifiers. Our experiments showed that AdaBoost makes a good performance in these two tasks.