Bridging the Gap Between Natural Language and Semantic Web: A Text-to-SPARQL System

<p>At the core of our research is a text-to-SPARQL query generation system that bridges natural language understanding with semantic data retrieval. The system leverages LLM agents with planning and Retrieval-Augmented Generation (RAG) capabilities alongside Wikidata API search to improve entity linking and efficiently retrieve candidate entities from Wikidata, while multiple Wikidata embedding strategies are explored to enhance retrieval accuracy. By automatically transforming user questions into precise SPARQL queries, our methodology ensures scalable extraction of information from large knowledge graphs. Multi-query ranking algorithms further refine the process, enhancing both the relevance and accuracy of retrieved results. Evaluation using the QALD dataset over multilingual Wikidata entities demonstrates the robustness and versatility of our solution, paving the way for more intelligent, user-friendly, and adaptable tools for navigating complex knowledge systems.</p>