Compound queries

Compound queries wrap or combine other queries to change their logic or score — bool, dis_max, function_score, boosting, and constant_score.

Why it matters

Real relevance tuning lives here. Beyond AND/OR, you often need to take the single best field instead of summing (cross-field search), boost by recency or popularity, demote (not exclude) certain docs, or strip scoring entirely. Each compound query is a precise tool for one of those shapes.

How it works

Each compound type composes child queries with different score arithmetic.

Query	Combines via	Use case
bool	AND/OR/NOT, additive	general logic
dis_max	max of child scores + tie_breaker	best single field wins
function_score	query × functions	boost by field value/decay/script
boosting	positive − negative_boost	demote without removing
constant_score	fixed score	turn a query into a filter

• dis_max — takes the highest child score so a strong title match isn’t diluted by weak body matches; tie_breaker (e.g. 0.3) adds a fraction of the others.
• function_score — multiply _score by field_value_factor, gauss/linear/exp decay (recency, geo distance), or a Painless script_score; combine with boost_mode/score_mode.
• boosting — positive matches normally, negative matches get multiplied by negative_boost (0–1) to sink them.
• constant_score wraps a filter, yielding filter context with a flat score.

Example

{ "function_score": {
    "query": { "match": { "title": "smartphone" } },
    "functions": [
      { "gauss": { "released": { "origin": "now", "scale": "30d", "decay": 0.5 } } },
      { "field_value_factor": { "field": "rating", "modifier": "log1p" } } ],
    "score_mode": "sum", "boost_mode": "multiply" } }

Newer, higher-rated phones rise: text relevance is multiplied by a recency decay and a dampened rating factor.

Pitfalls

• function_score order — boost_mode (query vs functions) vs score_mode (between functions) are distinct; confusing them flattens or explodes scores.
• script_score cost — runs per matching doc per shard; cache-unfriendly and slow on large result sets.
• dis_max without tie_breaker ignores secondary matches entirely — sometimes intended, often not.
• Decay scale units — scale:"30d" means score halves (at decay) 30 days out; mis-sizing it makes recency dominate or vanish.

See also

• bool-query-must-should-must-not-filter
• full-text-search