Topic model for personalized end-to-end task-oriented dialogue

摘要

Constructing a personalized end-to-end task-oriented dialogue system is one of the most important and challenging tasks in natural language processing technology. Slot-filling has achieved success in a rule-based task-oriented dialogue system. However, building a rule-based task-oriented dialogue system for real conversations is time-consuming. We present a novel personalized end-to-end framework based on split memory for Memory Networks by using topic model in this paper. We analyze the drawbacks of existing end-to-end dialog systems based on Memory Networks and propose the architecture which consists of user profile and conversation history. User profile is constructed by topic words from personalized topic model. The test experiments on the public and real dataset demonstrate that our method achieves better performance than the baselines in end-to-end task-oriented dialogue system.