# A Survey on Recent Advances in LLM-Based Multi-turn Dialogue Systems

**Authors:** Zihao Yi, Jiarui Ouyang, Yuwen Liu, Tianhao Liao, Zhe Xu, Ying Shen  
**arXiv:** [2402.18013](https://arxiv.org/abs/2402.18013)  
**Date:** 2024-02-28  
**Summary page:** https://www.summarizepaper.com/en/paper/2402.18013v1/

## Key points

- Comprehensive review of existing LLMs and methodologies for adapting them to different subtasks
- Detailed insights into state-of-the-art multi-turn dialogue datasets and evaluation metrics
- Discussion on future research directions and challenges in dialogue systems and LLM development

## Summary

This survey paper provides an in-depth overview of recent advancements in Large Language Model (LLM)-based multi-turn dialogue systems. The primary focus is on exploring the use of LLMs in dialogue systems and their adaptation to various downstream tasks. The paper aims to cater to a wide audience, including researchers and practitioners in academia and industry. The key contributions of this survey include: 
1. A comprehensive review of existing LLMs and methodologies for adapting them to different subtasks, along with an up-to-date analysis of LLM-based multi-turn dialogue systems. 2. Detailed insights into state-of-the-art multi-turn dialogue datasets and evaluation metrics. 3. Discussion on future research directions and challenges arising from the evolving demands on dialogue systems and the development of LLMs. The paper is structured as follows- Section 2 delves into general methods related to LLMs, highlighting their massive scale with billions of parameters and their effectiveness in learning intricate language representations. - Sections 3 to 4 provide a detailed exploration of methods for adapting LLMs to downstream tasks. - Section 5 covers important techniques for Task-Oriented Dialogue (TOD), including pipeline-based and end-to-end methods. - Section 6 presents cutting-edge methods for Open-Domain Dialogue (ODD) based on LLMs. - Sections 7 and 8 introduce relevant datasets and evaluation metrics for multi-turn dialogue systems. - Section 9 addresses key challenges and issues faced by LLM-based multi-turn dialogue systems. - Finally, the survey concludes in Section 10. Additionally, the comparison table in the content highlights different model structures used in popular LLM series such as GPT, BERT, T5, among others. This detailed summary aims to provide readers with a comprehensive understanding of the advancements in LLM-based multi-turn dialogue systems presented in this survey paper.