Hierarchical Multi-Agent Reinforcement Learning for Customer Service Dialogue Management

This repository provides the official implementation of the paper:

"Hierarchical Multi-Agent Reinforcement Learning for Robust and Scalable Customer Service Dialogue Management: A Decentralized Partially Observable Decision Perspective"

Overview

This project implements a hierarchical multi-agent reinforcement learning (HMARL) framework for AI-driven customer service dialogue management using a Dec-POMDP formulation.

DOI

A permanent archived version of this repository is available at:

https://doi.org/10.5281/zenodo.19061987

Related Paper

This repository corresponds to the manuscript:

"Hierarchical Multi-Agent Reinforcement Learning for Robust and Scalable Customer Service Dialogue Management: A Decentralized Partially Observable Decision Perspective"

If you use this code, please cite the paper.

Key Components

• Hierarchical manager-agent architecture
• Multi-objective reward optimization
• Coordination efficiency modeling
• Ablation study framework

Installation

pip install -r requirements.txt

Run

python train.py

Outputs

• Training reward curves
• Reward breakdown plots
• Coordination efficiency graphs
• Ablation comparison figures

Reproducibility

This implementation provides a reproducible simulation environment aligned with the methodology described in the paper.

Citation

If you use this code, please cite our paper.

Note

This repository is directly associated with a submission to Scientific Reports.