Boost match queries based on term proximity

[romseygeek]

Match queries currently score based on the standard BM25 bag-of-words model, where the input text is analyzed and terms are then added to a boolean query, using should or must clauses depending on the required query operator. This method does not take the positions of terms within the document into account. However, in many cases there is a correlation between the proximity of individual terms from the query and the relevance of the document.

We should add a proximity_boost parameter to the match query that takes a float value. Terms from the query can then be combined into an IntervalsSource.atLeast query that spans across each pairwise combination, giving a score bounded by 0 and 1 that gives a measure of the proximity of the terms, and this can then be converted to a boost as proximity_boost * (1 + interval_score) and applied to the query as whole via a function score.

[elasticmachine]

Pinging @elastic/es-search (:Search/Ranking)

[elasticsearchmachine]

Pinging @elastic/es-search (Team:Search)

[elasticsearchmachine]

Pinging @elastic/es-search-relevance (Team:Search Relevance)