Generation-Based and Emotion-Reflected Memory Update: Creating the KEEM Dataset for Better Long-Term Conversation

TL;DR

Long-term dialogue systems struggle to keep memory up-to-date as user state changes. The authors introduce KEEM, a generation-based memory-update approach that preserves essential information while embedding emotional context and causal relations, and they build KEEM by applying emotion–cause reflection and memory updating to Korean MSC dialogues using ChatGPT-4.0 with Korean few-shot prompts. Through extensive human and automatic evaluations, KEEM shows higher emotion–cause reflection, lower memory sentence conflicts, and more informative updates than prior methods (CareCallmem, KMSC), and yields better perplexity and response quality across multiple long-term chat models. This work provides a practical dataset and methodology for emotion-cognizant, causally grounded memory management in open-domain dialogue, with significant implications for empathy, personalization, and consistency in long-running conversations.

Abstract

In this work, we introduce the Keep Emotional and Essential Memory (KEEM) dataset, a novel generation-based dataset designed to enhance memory updates in long-term conversational systems. Unlike existing approaches that rely on simple accumulation or operation-based methods, which often result in information conflicts and difficulties in accurately tracking a user's current state, KEEM dynamically generates integrative memories. This process not only preserves essential factual information but also incorporates emotional context and causal relationships, enabling a more nuanced understanding of user interactions. By seamlessly updating a system's memory with both emotional and essential data, our approach promotes deeper empathy and enhances the system's ability to respond meaningfully in open-domain conversations.

Figures (5)

• Figure 1: Limitations of previous method and example of memory update.
• Figure 2: Process of KEEM dataset creation.
• Figure 3: Results of keyword coverage across the different memory update methodologies.
• Figure 4: Ratio of conflicts between memory sentences across the different memory update methodologies.
• Figure 5: Results of generation error analysis.