Multiplicative boosting scales a document’s complete relevance score by one or more functions. Unlike additive boosting, where feature scores are summed and each feature’s relative contribution decreases as features accumulate, multiplication lets signals scale the final score independently. section 3.2.3
Use multiplicative boosting to:
- Apply popularity, geographic distance, document age, field, category, phrase, or semantic signals without assigning every signal a fixed share of the total score. section 3.2.3 section 3.2.2
- Keep keyword relevance, such as BM25 matching, as the base calculation while multiplying it by a function such as 10 times a popularity field. section 3.2.3
- Choose between Solr’s boost query parser and the simplified edismax boost parameter when implementing the adjustment. section 3.2.3
{"query": "the cat in the hat",
"params": {"defType": "edismax",
"boost": "mul(popularity,10)"}}
{"query": "{!boost b=mul(popularity,10)} the cat in the hat"}
Compare equivalent Solr requests that multiply relevance by popularity, using either edismax’s boost parameter or the boost query parser. section 3.2.3
Popularity-scaled keyword search
For the query “the cat in the hat,” keyword matches can produce the base relevance score, while popularity independently scales that score through mul(popularity,10). A popular matching document can therefore outrank a less popular document with comparable keyword relevance. section 3.2.3
