Query-time boosting adjusts search ranking while processing each query. The system looks up signal-derived boosts in a separate collection, modifies the user’s query with those boosts, and then sends the boosted query to the search engine. section 8.6 section 8.6.1
A typical query-time boosting flow is:
- Receive an incoming user query. section 8.6.1
- Look up that query in a separate signals-boosting collection. section 8.6
- Use the returned document boosts to modify the user’s query. section 8.6.1
- Execute the modified query against the main search collection. section 8.6
Per-query document boosts
A signals-boosting lookup can return document identifiers with weights such as 966, 205, and 202. Those values are incorporated into the corresponding search query so selected documents receive stronger ranking influence.
The separation between the main collection and the boost collection supports:
- Incremental signal updates by changing only the document representing one query. section 8.6.1
- Easy activation or deactivation by skipping the lookup or query modification. section 8.6.1
- Experimentation with different signals-boosting algorithms and current-context ranking functions. section 8.6.1
This flexibility adds cost. Each request requires an extra search before the final search, increasing processing and latency. Limiting the number of boosted documents helps scale throughput but can reduce relevance. Pagination can also become inefficient or inaccurate when additional boosts change document positions across pages. section 8.6.1
"885909457588"^966 "885909457595"^205 "885909471812"^202 "886111287055"^109 "843404073153"^73 "885909457601"^62 "635753493559"^62 "885909472376"^61 "610839379408"^29 "884962753071"^28
Browse a compact listing of document identifiers and their query-time boost weights. section 8.6.1
