A Data-Driven Knowledge-Enriched Framework for Dialogue Modeling

In recent years, it has been advocated to build social dialogue systems to achieve human-machine interaction. In this study, we adopt deep learning models to develop a dialogue system. In contrast to other studies that mainly concentrate on the model structure and the learning method, our work emphasizes the importance of knowledge and explores the possibility of exploiting knowledge instead. Our system includes a procedure of training a deep learning model to generate answers in response to the users’ questions, and most importantly, a set of strategies to exploit various knowledge resources from different perspectives. We have conducted a series of experiments to evaluate the presented approach, and the results confirm its usefulness and effectiveness.