Memory-based Consultation System for Personalized Conversations Using Temporal Contexts

Due to the rapid development of natural language understanding technology, there are many attempts to apply interactive conversational systems in various domains such as consultation. In a conversational system, the quality of the generated dialogue has an important influence on the user's satisfaction, and its ability to lead a specialized conversation is required for the specific user. In this paper, we design a conversational system to manage user-specific memory information extracted from a conversation history corresponding to the user in the field of consultation. At the same time, the proposed system is configured to grasp the contextual information in the conversation history based on a language model learned to extract temporal contexts. We utilize the contexts and keywords to transform the major contents of the conversation history into episodic memory. By accumulating the episodic memories for the particular user and structuring them as long-term memory, we develop the basis to provide personalized conversation to the user. Based on the memory built for each user, we believe that customization can be possible depending on the user's historical information in the process of dialogue generation.