Personalized Conversation Management with Emotional Memory

Due to rapid advancements in natural language comprehension technologies, numerous efforts have been undertaken to implement interactive conversational systems across various fields, including counseling services. In a conversational system, the caliber of the generated dialogue significantly impacts user satisfaction, necessitating the ability to navigate specialized discussions tailored to individual users. In this paper, we propose a conversational system designed with a user-centric approach to effectively manage the emotional dimensions of user-specific memory data derived from conversational histories. Simultaneously, the proposed system comprehends and utilizes contextual information from the conversation history, leveraging a language model prompted to extract emotional contexts specifically. We also employ identified contexts and key terminologies to convert the primary elements of conversation history into an emotional memory framework. By aggregating each user’s emotional memories and organizing them as a form of long-term memory, we establish a foundational basis for delivering personalized conversational experiences. Based on the memory frameworks established for each user, we believe personalization is achievable depending on the user’s historical data throughout the dialogue generation process with emotional consideration.