RANGER: Retrieval Agent for Navigating Graphs with Semantic Evaluation Rewards

Key points

• This paper introduces a novel hybrid approach in which an LLM serves as a mentor to an RL agent during training by providing dense, semantically informed reward signals.1
• The RL agent, built on a Transformer-based policy and trained with Group Relative Policy Optimization, learns to navigate a knowledge graph to retrieve relevant information.1
• We integrate this Retrieval Agent for Navigating Graphs with Semantic Evaluation Rewards (RANGER) into a complete question-answering pipeline and evaluate it on the open question answering datasets.1
• Experimental results show that the proposed approach outperforms both a baseline retrieval-augmented system with classical graph search and one using an RL agent trained with heuristic rewards, achieving a significant increase in NLP metrics and LLM-as-a-Judge scores.1

Sources (1)

• [1]RANGER: Retrieval Agent for Navigating Graphs with Semantic Evaluation Rewards
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 10:53 AM

  The integration of Large Language Models and Reinforcement Learning offers a promising direction for complex sequential decision-making tasks, such as multi-hop question answering over knowledge graphs.

  This paper introduces a novel hybrid approach in which an LLM serves as a mentor to an RL agent during training by providing dense, semantically informed reward signals.

Extractive summary: sentences quoted from the sources.