Design of pre-school education chatbot based on data-driven and deep reinforcement learning
In order to enable pre-school education chatbots to provide more personalized dialogue interaction services,a dialogue strategy model based on user portrait and deep reinforcement learning is proposed.Firstly,the history state of the dialogue action is modeled through the gated loop unit,and the action history vector containing the user behavior characteristics is extracted.Then the extracted action history vector is combined with the user portrait vector and the current conversation state vector to input the action val-ue network.Finally,through the action value network,the model can find the best response action that is most in line with the current user's personality,and generate the corresponding best dialogue strategy.The experimental results show that the performance of the proposed model is best when the parameter k of the dialog action history window is 3.Compared with the dialogue strategy models based on deep reinforcement learning commonly used in current dialogue systems such as DQN,DRQN and Dueling,the proposed model achieves 0.45,17.93 and 26.22,respectively,in terms of the dialogue success rate,average dialogue reward and average number of dialogue rounds,with obvious improvement effect and better dialogue quality and efficiency.It is worth further populariza-tion and research.
deep reinforcement learninguser portraitpreschool educationdialogue interactionneural network
万方数据
维普
