Emotion Analysis for the Upcoming Response in Open-Domain Human-Computer Conversation

Emotion analysis is one of the most active domains, hence attracts lots of attention of researchers in the natural language processing field. However, most of existed works are involved in classification tasks of the current sentence, lack of analysis of upcoming sentences. On the other hand, with the development of automatic human-computer dialogue systems, a response given by the computer side should become increasingly like human beings, for instance, the ability of expressing sentiment or emotion. The challenges lies in how to predict the emotion of a nonexistent sentence currently, which make this problem quite different from traditional sentiment or emotion analysis. In this paper, for the scenarios of open-domain conversation, we propose an architecture based on deep neural networks to predict the emotion before giving the response. In particular, we use a bidirectional recurrent neural network to get the embedding of the current utterance, and joint the representations of its retrieval results, to obtain the best emotion classification of the upcoming response. Experiments based on an annotation dataset demonstrate the effectiveness of our proposed approach better than traditional methods in terms of accuracy, precision, recall, and F-measure evaluation metrics. Then the following is some analysis of the results and future works.