Information retrieval
The information-science subfield concerned with searching for documents, for information within documents, and for metadata about them information-science-wikipedia. Automated IR exists to answer information overload; its span runs from a library catalogue to a web search engine.
The basic shape
A user enters a query; the system matches it against a collection; results come back ranked by a relevance score rather than as a set. Ranking is the part that makes IR its own discipline — a Boolean match tells you whether a document qualifies, not which of ten thousand qualifying documents to read first. Boole's 1854 algebra is named in the same article as the ancestor of the query side.
Why it belongs in this wiki
The corpus's largest technical cluster is retrieval: retrieval-augmented-generation, hybrid-retrieval-rag, graphrag, lightrag, hipporag2, leanrag, graphrag-bench, long-context-vs-chunking. Every one of them is an IR system with a language model attached, and the corpus has been discussing them without the parent field's name on any page.
Two things the parent field supplies that the RAG cluster mostly does not:
- Relevance is a judgment, not a property. IR evaluation is built on human relevance assessments, with the disagreement between assessors treated as part of the measurement. graphrag-bench is the corpus's closest equivalent, and it scores task outcomes rather than retrieval relevance.
- The searcher is a research subject. Information seeking behaviour — how professionals actually look for things — is a named subfield with its own models. Nothing in this corpus measures the person; every benchmark here measures the pipeline.
Access is treated as a neighbouring area: text mining, categorization, machine translation, and the privilege/restriction layer that decides who may retrieve what.
Related
information-science · retrieval-augmented-generation · hybrid-retrieval-rag · graphrag · knowledge-graph · bibliometrics · documentation-movement