GraphRAGProv tracks sources to explain graph query answers
GraphRAGProv: GraphRAG Meets How-Provenance
Artificial IntelligenceDatabases
Summary
Sometimes computers answer questions by looking up connected facts in a big graph, but it can be hard to see how they got the answer. The authors made a new system called GraphRAGProv that not only finds answers but also shows which exact facts were used and where those facts came from. This way, people can check if the answers really come from the original texts and understand how the computer combined the facts. Testing the system found mistakes in how facts were pulled or combined, even when the final answer was correct.
What this means in practice
- •For technical support teams: Provide clear traceable explanations for automated answers to improve customer trust and verification in complex knowledge domains.
- •For biomedical data teams: Inspect and validate extracted biomedical literature facts and their combinations to ensure transparent automated queries for research data.
Authors
Wadie El Amrani, Luis Galárraga
Abstract
GraphRAG systems often use knowledge graphs to retrieve relevant context for an LLM that produces the final answer. That makes answers more inscrutable and therefore less appealing for domains where transparency is paramount. We present GraphRAGProv, an architecture that executes SPARQL queries over source-grounded RDF graphs to compute answers and their how-provenance polynomials. These polynomials record which statement occurrences were used and how they were combined. Because each occurrence is linked to its source text, users can follow these links to check whether the extracted statements reflect the text and whether the query combines them to answer the question. Querying thereby becomes a way to inspect and evaluate the text-to-graph transformation. Our interactive demo examines 400 cases from two multi-hop QA benchmarks, revealing extraction and query errors even when the final answer is correct.