Optimizing Of Dialogue Policy In Human-computer Spoken Dialogue System Based On Reinforcement Learning

Spoken Dialogue System is concerned with the development of Artificial Intelligence, Cognitive Science and Linguistics. The development of Natural Language Understanding made it possible to apply spoken dialogue system to many domains. This will change the way of human-computer interaction and our life.Reinforcement Learning is one of important machine learning methods, which has the characteristic of self-improving. Reinforcement Learning regards learning as a process of trial and enor. States in environment are mapped to actions in Reinforcement Learning. It can solve the problem very well, which is how an agent searches the best action in random state environment.Reinforcement Learning is applied in Spoken Dialogue System in this thesis, the model of Spoken Dialogue System presented by Singh and Walker is improved. The improved model named RL-SDS (Reinforcement Learning-Spoken Dialogue System) has better applicability. RL-SDS model is applied in two spoken dialogue environments-computer practice query system and psychology consultation system. The thesis has a comparison to RL-SDS model and Singh's model under the spoke dialogue environment of computer practice query system. Singh and Walker's model is not capable of handling the spoke dialogue environment of psychology consultation system, whereas RL-SDS model has an analysis to effect of Reinforcement Learning.