MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202641110301 A) filed by Vellore Institute Of Technology on September 14, 2026, for System And Method For Provenance-Based Correction Of Machine-Extracted Knowledge Graphs Using A Closed Trust-Feedback Loop.
Inventors include Vani Rajasekar; and Vaibhav Pandey.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: The present invention provides a provenance-aware knowledge graph correction system and method for selectively repairing machine-extracted facts without re- ingestion, recrawling, or model retraining. Facts extracted from source documents are stored as non-destructive reified claims associated with corresponding entities, properties, and originating sources having trust values. Conflicting claims are deterministically arbitrated using source trust, recency, and extraction confidence, while non-winning claims are retained as superseded records. A query-time verification outcome for a served answer is traced through its provenance chain to the originating source, upon which a closed-form trust update is applied. A source index then identifies other claims contributed by the source, including previously unserved facts, for selective re-arbitration. Trust evolution is recorded in an append-only replayable event log. The system further provides entity resolution, corroboration-based confidence aggregation, query-time ranking, and confidence-based gating or abstention, thereby providing deterministic, non-destructive, and computationally efficient knowledge graph correction.
Disclaimer: Curated by HT Syndication.
