Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Conversational Intent-Driven GraphRAG: Enhancing Multi-Turn Dialogue Systems through Adaptive Dual-Retrieval of Flow Patterns and Context Semantics

Ziqi Zhu, Tao Hu, Honglong Zhang, Dan Yang, HanGeng Chen, Mengran Zhang, Xilun Chen

TL;DR

This work presents CID-GraphRAG, a framework that integrates dynamic conversational intent-transition graphs with semantic retrieval to address the dual challenges of maintaining contextual coherence and goal-oriented progression in knowledge-intensive, multi-turn customer service dialogues. It introduces a construction phase that automatically builds an intent graph from goal-achieved historical conversations and an inference phase that performs dual-pathway retrieval, balancing intent-pattern traversal with semantic context via adaptive weighting. The approach yields a structured prompt for generation grounded in both conversation structure and context, leading to superior retrieval and response quality across automatic metrics, LLM-based judgments, and human evaluation, compared with Conversation RAG and GraphRAG baselines. The results demonstrate significant improvements, including a 58% increase in LLM-as-judge response quality, and point to practical gains in real-world deployment, with future work on adaptive intent hierarchies, reinforcement learning, and cross-domain applicability.

Abstract

We present CID-GraphRAG (Conversational Intent-Driven Graph Retrieval Augmented Generation), a novel framework that addresses the limitations of existing dialogue systems in maintaining both contextual coherence and goal-oriented progression in multi-turn customer service conversations. Unlike traditional RAG systems that rely solely on semantic similarity (Conversation RAG) or standard knowledge graphs (GraphRAG), CID-GraphRAG constructs dynamic intent transition graphs from goal achieved historical dialogues and implements a dual-retrieval mechanism that adaptively balances intent-based graph traversal with semantic search. This approach enables the system to simultaneously leverage both conversional intent flow patterns and contextual semantics, significantly improving retrieval quality and response quality. In extensive experiments on real-world customer service dialogues, we employ both automatic metrics and LLM-as-judge assessments, demonstrating that CID-GraphRAG significantly outperforms both semantic-based Conversation RAG and intent-based GraphRAG baselines across all evaluation criteria. Quantitatively, CID-GraphRAG demonstrates substantial improvements over Conversation RAG across automatic metrics, with relative gains of 11% in BLEU, 5% in ROUGE-L, 6% in METEOR, and most notably, a 58% improvement in response quality according to LLM-as-judge evaluations. These results demonstrate that the integration of intent transition structures with semantic retrieval creates a synergistic effect that neither approach achieves independently, establishing CID-GraphRAG as an effective framework for addressing the challenges of maintaining contextual coherence and goal-oriented progression in knowledge-intensive multi-turn dialogues.

Conversational Intent-Driven GraphRAG: Enhancing Multi-Turn Dialogue Systems through Adaptive Dual-Retrieval of Flow Patterns and Context Semantics

TL;DR

This work presents CID-GraphRAG, a framework that integrates dynamic conversational intent-transition graphs with semantic retrieval to address the dual challenges of maintaining contextual coherence and goal-oriented progression in knowledge-intensive, multi-turn customer service dialogues. It introduces a construction phase that automatically builds an intent graph from goal-achieved historical conversations and an inference phase that performs dual-pathway retrieval, balancing intent-pattern traversal with semantic context via adaptive weighting. The approach yields a structured prompt for generation grounded in both conversation structure and context, leading to superior retrieval and response quality across automatic metrics, LLM-based judgments, and human evaluation, compared with Conversation RAG and GraphRAG baselines. The results demonstrate significant improvements, including a 58% increase in LLM-as-judge response quality, and point to practical gains in real-world deployment, with future work on adaptive intent hierarchies, reinforcement learning, and cross-domain applicability.

Abstract

We present CID-GraphRAG (Conversational Intent-Driven Graph Retrieval Augmented Generation), a novel framework that addresses the limitations of existing dialogue systems in maintaining both contextual coherence and goal-oriented progression in multi-turn customer service conversations. Unlike traditional RAG systems that rely solely on semantic similarity (Conversation RAG) or standard knowledge graphs (GraphRAG), CID-GraphRAG constructs dynamic intent transition graphs from goal achieved historical dialogues and implements a dual-retrieval mechanism that adaptively balances intent-based graph traversal with semantic search. This approach enables the system to simultaneously leverage both conversional intent flow patterns and contextual semantics, significantly improving retrieval quality and response quality. In extensive experiments on real-world customer service dialogues, we employ both automatic metrics and LLM-as-judge assessments, demonstrating that CID-GraphRAG significantly outperforms both semantic-based Conversation RAG and intent-based GraphRAG baselines across all evaluation criteria. Quantitatively, CID-GraphRAG demonstrates substantial improvements over Conversation RAG across automatic metrics, with relative gains of 11% in BLEU, 5% in ROUGE-L, 6% in METEOR, and most notably, a 58% improvement in response quality according to LLM-as-judge evaluations. These results demonstrate that the integration of intent transition structures with semantic retrieval creates a synergistic effect that neither approach achieves independently, establishing CID-GraphRAG as an effective framework for addressing the challenges of maintaining contextual coherence and goal-oriented progression in knowledge-intensive multi-turn dialogues.

Paper Structure

This paper contains 30 sections, 2 equations, 4 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. System Overview
  5. Construction Phase
  6. Intent Recognition
  7. Graph Construction
  8. Inference Phase
  9. Dual-Pathway Retrieval
  10. Adaptive Integration
  11. Response Generation
  12. Experiments
  13. Experimental Setup
  14. Dataset
  15. Evaluation Metrics
  16. ...and 15 more sections

Figures (4)

  • Figure 1: The detailed framework of CID-GraphRAG. The CID-GraphRAG consists of two phases: (1) a construction phase that builds an intent graph from goal-achieved conversations, and (2) an inference phase that identifies user and assistant intents from current dialogue turn, retrieves high-quality examples from both intent-based and semantic-based pathway via adaptive weighting mechanism, and uses LLM for structured response generation.
  • Figure 2: Internal structure of the CID-Graph.The graph comprises distinct primary intent and secondary intent nodes and conversation nodes. Key relations include hierarchical, pairing, transition, and dialogue anchoring.
  • Figure 3: Comparison between semantic matching and exact matching on 58 cases. Semantic matching outperforms exact matching across all weight configurations in CID-GraphRAG.
  • Figure 4: Performance comparison of different methods based on LLM-as-Judge win counts. The left group shows retrieval quality wins, and the right group shows response generation wins. CID-GraphRAG consistently outperforms all baseline methods in both aspects, with particularly significant advantages in retrieval quality.