Follow the Entities: A Corpus Map for Agentic Search
CorpusMap organizes document collections around recurring entity pages so agents recover cross-document evidence using fewer tokens.
CorpusMap is an offline navigation layer that resolves recurring entities across a document collection and represents each as an Entity Page linked to every document that mentions it. Agents can traverse that entity-document graph to connect evidence that a flat file search would rediscover query by query. Across seven models and three benchmarks, CorpusMap improved evidence discovery and answer quality versus raw-corpus agentic search while using fewer tokens on average, and it beat four alternative navigation layers.
- CorpusMap builds offline Entity Pages linking recurring entities to every mentioning document.
- Agents traverse an entity-document graph instead of rediscovering relations per query.
- Across seven models and three benchmarks it improved answers while using fewer tokens.
- It outperformed four alternative navigation layers for large corpora.
From the original article
1 min read at huggingface.co
Answering questions and completing tasks over large document collections often requires connecting evidence spread across multiple documents, such as a project's approval recorded in one, its requirements in another, and its latest status in a third. Recent LLM agents approach this by iteratively searching the full corpus rather than reading only a fixed set of top-ranked documents.
By Soyeong Jeong, Sujay Kumar Jauhar, Sung Ju Hwang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.