非タスク指向型対話システムの改善

抄録

With the development of information and communication technology, various technologies and services have been born. Among them, the dialogue system is one of the technologies that have received a lot of attention. In this study, proposed non-task-oriented dialogue system is using Recurrent Neural Network (RNN) and word embedding. In dialogue systems, there are two types. One is retrieval model and the other is generative model. The dialogue system using retrieval model has few mistakes of grammar. However, it is like a parrot and takes a few seconds to generate responses. Therefore, we propose the dialogue system using generative model to improve these problems. In the proposed system, method of generating responses is seq2seq. The seq2seq is a RNN model that is trained by training data that pairs of inputs and outputs. Furthermore, the proposed system was improved to use enough vocabulary using Word2Vec. The proposed system can generate responses more natural than the retrieval model. Also, response speed was improved. Although some improvements were found, the proposed system was shown to be suitable as a dialogue system.