Relevance

Relevance is the connection between topics that makes one useful for dealing with the other. It is studied in cognitive science, logic, and library and information science, with epistemology examining its general nature.

Key measured quantities

Practical procedure

1. Identify the user's information need and express it as a query.
2. Retrieve candidate documents or items using an index such as an inverted index.
3. Compute a relevance score for each candidate using a ranking function like BM25 or TF-IDF.
4. Rank the candidates by descending relevance score.
5. Present the top-ranked results to the user.
6. Collect user feedback (e.g., clicks) to adjust future rankings.

Information Retrieval

In information retrieval, relevance measures how well a document meets a query. The TREC conference, started in 1992, established precision and recall as standard metrics. For example, a system might achieve 80% precision at rank 10.

Library and Information Science

Libraries use relevance to organize collections and assist patrons. The Dewey Decimal Classification, introduced in 1876, groups related subjects. A library might report a 90% satisfaction rate for relevant search results.

Cognitive Science

Cognitive science studies how humans judge relevance. The ACT-R theory, developed by John R. Anderson in 1976, models memory retrieval. Experiments show that relevant information is recalled 30% faster than irrelevant information.

Logic and Epistemology

In logic, relevance logic requires premises to be relevant to conclusions. The system R, formulated by Anderson and Belnap in 1975, avoids paradoxes of material implication. A proof in relevance logic may require 5 steps instead of 2.

Questions visitors ask

What is the difference between precision and recall?

Precision is the fraction of retrieved documents that are relevant, while recall is the fraction of relevant documents that are retrieved. For a query with 10 relevant documents, retrieving 5 relevant and 5 irrelevant gives precision 0.5 and recall 0.5.

How is relevance measured in search engines?

Search engines use metrics like mean reciprocal rank (MRR) and normalized discounted cumulative gain (NDCG). A perfect MRR is 1.0, meaning the first result is always relevant.

What is relevance logic?

Relevance logic is a non-classical logic that requires the antecedent to be relevant to the consequent. It was formalized in the 1970s by Anderson and Belnap, with systems like R and E.