Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces MARS, a novel approach to knowledge graph question answering (KGQA) that integrates large language models (LLMs) with knowledge graphs (KGs) without requiring model fine-tuning. MARS employs a structured retrieval process that iteratively links question entities to the KG, allowing it to adaptively decide on the depth of retrieval or generate a SPARQL query based on the question's complexity. The evaluation across multiple benchmarks demonstrates that MARS achieves competitive performance against state-of-the-art methods while maintaining efficiency and scalability, making it a promising solution for knowledge-intensive tasks.
MARS adapts retrieval depth dynamically, achieving state-of-the-art KGQA performance without the overhead of model fine-tuning.
Large language models (LLMs) have demonstrated strong reasoning performance, but their tendency to hallucinate limits their reliability in knowledge-intensive tasks requiring up-to-date and grounded information. Combining knowledge graphs (KGs) with LLMs facilitates the use of explicit symbolic knowledge that can be continuously updated without costly fine-tuning, while benefiting from rapidly advancing LLM reasoning. We propose MARS, a scalable knowledge graph question answering (KGQA) approach that requires no model fine-tuning. Rather than relying on open-ended agentic exploration, MARS performs a structured retrieval procedure that links question entities to the KG and iteratively retrieves relevant next-hop information. At each step, MARS decides whether to continue graph traversal or to generate the final SPARQL query, allowing the model to adapt the retrieval depth to the question while keeping the overall pipeline more predictable than fully agentic approaches. We evaluate MARS on three established KGQA benchmarks across several LLMs and settings, including multilingual evaluation, and provide insights through ablation studies and error analysis. Our approach achieves competitive performance relative to state-of-the-art methods while remaining efficient and scalable. The evaluation results, code and resources are publicly available: https://github.com/dice-group/mars-kgqa.