Minimum Should Match

Boolean-Retrieval Query-Semantics Disjunctive-Clauses Needs-Review

What it is

In boolean queries, MUST clauses require matching; MUST NOT clauses forbid matching. SHOULD clauses are optional, improving score if matched. Minimum should match (MM) specifies the minimum number of SHOULD clauses that must match for a document to be retrieved.

How it works

Boolean query execution:

  1. MUST clauses: document must match all
  2. SHOULD clauses: optional; document matches if ≥ MM of them match
  3. MUST NOT clauses: document must not match any
  4. Scoring: MUST clauses contribute fixed importance; SHOULD matches increase score proportionally

MM can be specified as:

  • Absolute count: “2” (at least 2 SHOULD clauses)
  • Percentage: “75%” (at least 75% of SHOULD clauses)
  • Expression: “1<-25% 9<-3” (1 below 9 terms: 25% of clauses; 9 or more: 3 clauses)

[illustrate: Query with multiple SHOULD clauses and varying MM thresholds showing which documents match]

Example

Query: (lucene OR elasticsearch OR solr) AND language:java

  • MM=1: documents with java and any one search engine
  • MM=2: documents with java and any two search engines
  • MM=3: documents with java and all three search engines

Variants and history

Lucene and Elasticsearch both support MM configuration. Useful for precision tuning in complex disjunctive queries. Sometimes called “minimum match” or “match clauses”. Related to fuzzy matching (k of n constraints).

When to use it

Control recall-precision in complex queries with many SHOULD clauses. Use MM=1 for broad recall; increase MM to narrow results. Useful when query contains multiple variations or optional conditions. Allows query-time fine-tuning without reformulation.

See also