We introduce EXYGEN (Explore Your Graphs Engine), a framework for knowledge graph (KG) understanding that enables conversational access at scale. First, we evaluate how well large language models (LLMs) can perform text‑to‑SPARQL generation when only automatically derived structured metadata and small graph samples are available, without task‑specific fine‑tuning. VoID descriptions and ShEx schemas are integrated into a retrieval‑augmented generation (RAG) pipeline, and KG‑derived context is ablated on the SciQA benchmark. The best configuration—combining ShEx schemas, retrieved triples, and example question‑query pairs—achieves an exact match of 0.419 on execution results without any LLM fine‑tuning. Experiments show that lexical metrics such as F1 poorly predict query correctness, and larger general‑purpose LLMs can outperform smaller code‑specialized models when given sufficient context. Second, we address the computational bottleneck of generating metadata for very large KGs by proposing a predicate‑coverage‑aware parallel graph sampling strategy that preserves structural diversity while remaining tractable. On OpenCitations Meta and GESIS it retains high predicate coverage with minimal triple loss and reduces runtime by over 80×; on ORKG sampling is not only faster but the only feasible way to obtain complete metadata. Together, these results demonstrate that structured schema context and lightweight prompting can substantially reduce reliance on fine‑tuning for scalable conversational KG access. Closing the remaining gap to fully fine‑tuned approaches will likely require reducing dependence on curated question‑query exemplars, either via synthetic generation or execution‑feedback‑driven methods, and validating the findings across additional benchmarks.
Review