A Dialogue System with Emotion Estimation and Knowledge Acquisition Functions

Abstract

In this paper, we propose a dialogue system with emotion estimation and knowledge acquisition functions. In the emotion estimation part, 9-emotions are treated in the proposed bimodal approach using voice and natural language processing. In the voice processing part, 396-dimensional features are extracted from the voice data, and they are used in learning of Stacked Denoising AutoEncoder. In the natural language processing part, two methods are proposed. The first one is to consider the implications of the relationship between emotional expression dictionary and input sentence's words. The second one is to consider the co-occurrence frequency of the emotion words in emotion dictionary. In the knowledge acquisition part, the relationship patterns from the input sentences are extracted. In the evaluation experiments, we can confirm that the proposed dialogue system obtains higher score than the existing system in term of the adequacy of the feeling estimation and the variety of response sentences.