Follow the Entities: A Corpus Map for Agentic Search
CorpusMap gives search agents an entity graph to follow instead of making them rediscover document links query by query.
The paper proposes building “Entity Pages” offline from recurring entities found across a document collection. Those pages aggregate mentions and link back to every document that refers to the same entity, giving agents a navigation layer between otherwise flat files. In tests across seven models and three benchmarks, the authors report better evidence discovery and answer quality than raw-corpus agentic search, with fewer tokens on average. HF Daily Papers' note
The paper proposes building “Entity Pages” offline from recurring entities found across a document collection. Those pages aggregate mentions and link back to every document that refers to the same entity, giving agents a navigation layer between otherwise flat files. In tests across seven models and three benchmarks, the authors report better evidence discovery and answer quality than raw-corpus agentic search, with fewer tokens on average. HF Daily Papers' note
score 4