Compress to Impress: Unleashing the Potential of Compressive Memory in Real-World Long-Term Conversations

TL;DR

This paper introduces COMEDY, a novel compressive memory framework for dialogue systems that manages long-term conversations more effectively by using a single language model to generate, compress, and utilize memory without traditional retrieval modules.

Contribution

The paper proposes a new compressive memory approach called COMEDY that simplifies long-term dialogue management by eliminating retrieval modules and using a unified language model.

Findings

COMEDY outperforms traditional retrieval-based methods in nuanced conversation quality.

The curated Dolphin dataset supports large-scale Chinese instruction tuning.

COMEDY achieves more human-like and dynamic conversational responses.

Abstract

Existing retrieval-based methods have made significant strides in maintaining long-term conversations. However, these approaches face challenges in memory database management and accurate memory retrieval, hindering their efficacy in dynamic, real-world interactions. This study introduces a novel framework, COmpressive Memory-Enhanced Dialogue sYstems (COMEDY), which eschews traditional retrieval modules and memory databases. Instead, COMEDY adopts a "One-for-All" approach, utilizing a single language model to manage memory generation, compression, and response generation. Central to this framework is the concept of compressive memory, which intergrates session-specific summaries, user-bot dynamics, and past events into a concise memory format. To support COMEDY, we curated a large-scale Chinese instruction-tuning dataset, Dolphin, derived from real user-chatbot interactions. Comparative evaluations demonstrate COMEDY's superiority over traditional retrieval-based methods in producing more nuanced and human-like conversational experiences. Our codes are available at https://github.com/nuochenpku/COMEDY.