Select, Don't Train: The Benefits of Modular Entity Disambiguation with LLM-Based Selection
Source
arxiv.org
Author
Fina Polat, Daniel Daza, Pengyu Zhang, Klim Zaporojets, Paul Groth
Date
Context
Matching a name in text to the right knowledge-base entry can be split into finding plausible candidates and choosing among them using context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition →. The Select, Don't Train paper tests this division by pairing several retrieval methods with a LLMA large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.Full definition → that makes the final selection.
On the ZELDA benchmarkA standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.Full definition →, BM25 term matching followed by OpenAI's GPT-5.4-mini scored 86.3 on the authors' inKB micro-F1 measure, compared with 82.3 for the trained VERBALIZED retriever used end to end. A trained retriever paired with the same selector reached 88.5. The result makes simple retrieval a credible starting point while showing that specialized retrieval can still improve accuracy.
Terms in this piece · Glossary
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
articleFrom Pixels to Pairs: A Comprehensive Benchmark of LLM-Based Key-Value Extraction in Noisy Document SettingsZahra Anvari, Vassilis Athitsos
articleAn Ontology-Guided, Deduplication-Aware Extraction Layer for Knowledge Graph Construction from Heterogeneous DocumentsVaibhav Dangaich, Kevin Lewis, Kundeshwar Pundalik
articleAgentic Data Engineering for LLMs: System Design and a Controlled Empirical StudyRuomeng Ding, Qianli Shen, Zhaoyang Han, Meixin Chen, Daoyuan Chen, Yaliang Li