An Ontology-Guided, Deduplication-Aware Extraction Layer for Knowledge Graph Construction from Heterogeneous Documents
Source
arxiv.org
Author
Vaibhav Dangaich, Kevin Lewis, Kundeshwar Pundalik
Date
Why it matters
Injecting only the relevant ontology slice per document cuts prompt catalog overhead ~94% while pushing search recall from about 70% to 95%, and the conflict guard shows how to stop embeddingA list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.Full definition → from silently merging two people with the same name.
Terms in this piece · Glossary
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
articleSelect, Don't Train: The Benefits of Modular Entity Disambiguation with LLM-Based SelectionFina Polat, Daniel Daza, Pengyu Zhang, Klim Zaporojets, Paul Groth
articleNexus: Depth-Adaptive KV-Cache Splicing and Retrieval-Decoupled Tool Routing for Agentic LLMs on Unified MemoryMustafa Arslan
articleFrom Pixels to Pairs: A Comprehensive Benchmark of LLM-Based Key-Value Extraction in Noisy Document SettingsZahra Anvari, Vassilis Athitsos