Signal aggregation combines recorded search interactions into query-document measures. A typical pipeline joins query events to clicked documents, counts interactions, and stores the resulting boosts in a sidecar collection for query-time ranking. section 4.2.2
A practical aggregation pipeline:
- Join each interaction to its originating query and document. section 4.2.2
- Group by query and document, then count or weight the interaction events. section 4.2.2 section 8.4
- Normalize equivalent query forms before aggregation so their evidence is combined instead of fragmented. section 8.2
- Write the aggregated values to a sidecar collection and use them as query-time boosts. section 4.2.2
Weighted signals for an iPad query
For a query such as "ipad", a document receives one point per click, ten per add-to-cart event, and twenty-five per purchase. The resulting query-document score can be expressed as (1 × clicks) + (10 × add-to-carts) + (25 × purchases). A ranked output might give one document a boost of 966 and another 205, then convert those values into document boosts so selected products rise in search results. section 8.4
