ARTFEED — Contemporary Art Intelligence

MATRAG: Multi-Agent Framework for Explainable LLM Recommendations

ai-technology · 2026-04-25

MATRAG, which stands for Multi-Agent Transparent Retrieval-Augmented Generation, is designed to improve how recommendation systems using large language models handle transparency and provide explanations. The system features four unique agents: one that adjusts user preference profiles, another that analyzes items by pulling semantic details from knowledge graphs, a reasoning agent that combines collaborative and content signals, and an explanation agent that creates understandable justifications in plain language. By bringing together these agents and leveraging knowledge graphs for better retrieval, the goal is to deliver recommendations that users can trust, thanks to clear, knowledge-driven explanations.

Key facts

  • MATRAG stands for Multi-Agent Transparent Retrieval-Augmented Generation
  • The framework uses four specialized agents: User Modeling, Item Analysis, Reasoning, and Explanation
  • It combines multi-agent collaboration with knowledge graph-augmented retrieval
  • The User Modeling Agent constructs dynamic preference profiles
  • The Item Analysis Agent extracts semantic features from knowledge graphs
  • The Reasoning Agent synthesizes collaborative and content-based signals
  • The Explanation Agent generates natural language justifications
  • The framework addresses transparency, knowledge grounding, and explanation challenges

Entities

Sources