KBQA-R1: Reinforcing Large Language Models for Knowledge Base Question Answering

The article introduces KBQA-R1, a framework designed to enhance Knowledge Base Question Answering (KBQA) by employing Reinforcement Learning to optimize interaction with knowledge bases. It utilizes Group Relative Policy Optimization (GRPO) for strategy refinement based on execution feedback and introduces Referenced Rejection Sampling (RRS) to align reasoning with ground-truth actions, addressing cold-start issues. Extensive experiments show that KBQA-R1 achieves state-of-the-art performance on benchmarks like WebQSP, GrailQA, and GraphQuestions, making it a significant advancement for practitioners seeking to improve LLM reasoning in verifiable environments.