liveBook · Manning

Click signal

As explained in AI-Powered Search

A click signal is a recorded event indicating that a user clicked a search result document associated with a query.

A click signal records that a user clicked a search-result document for a particular query. Search systems can aggregate these events to estimate which documents users find relevant and boost them in later rankings. section 4.1.3 section 8.4

Represent each click with a query identifier, user, signal type, target document, and timestamp. The query identifier links the click to the query that produced the result, allowing aggregation by query and document. section 4.1.3

Use click signals in a broader relevance pipeline:

  • Count clicks for each query/document pair to create a popularity or relevance boost. section 8.3.1
  • Combine clicks with add-to-cart and purchase events when stronger evidence of product interest is available. section 8.4
  • Assign different weights to reduce the influence of weaker signals; one demonstrated configuration gives clicks weight 1, add-to-cart events 10, and purchases 25. section 8.4

Worked interaction sequence

A user searches for “ipad,” clicks doc1 and doc3, and adds doc3 to a cart. The query identifier connects both clicks to “ipad,” while separate timestamps preserve the interaction order. A later “ipad cover” query links a click and cart addition for doc4, followed by purchases of doc3 and doc4. section 4.1.3

Clicks are useful but noisy: browsing can produce a click without strong purchase intent. They are also vulnerable to manipulation, such as repeatedly querying “star wars” and clicking one product thousands of times, which can push that product to the top. section 8.4 section 8.3.1

Table 4.1 Example signals format
query_id
user
type
target
signal_time
1
u123
query
ipad
2024-05-01-09:00:00
1
u123
results
doc1,doc2,doc3
2024-05-01-09:00:00
1
u123
click
doc1
2024-05-01-09:00:10
1
u123
click
doc3
2024-05-01-09:00:29
1
u123
add-to-cart
doc3
2024-05-01-09:03:40
2
u123
query
ipad cover
2024-05-01-09:04:00
2
u123
results
doc4,doc5
2024-05-01-09:04:00
2
u123
click
doc4
2024-05-01-09:04:40
2
u123
add-to-cart
doc4
2024-05-01-09:05:50
1
u123
purchase
doc3
2024-05-01-09:07:15
2
u123
purchase
doc4
2024-05-01-09:07:15

Inspect the tabular event representation linking queries, returned results, clicks, cart additions, and purchases. section 4.1.3

How AI-Powered Search covers it

How deep it goes
Defines the event schema and weighted aggregation, then exposes spam vulnerability without detailing defenses.
Examples
Uses an ipad shopping journey, weighted SQL aggregation, and star-wars click spam to illustrate collection and failure.
What it assumes
Assumes search queries and returned documents, while teaching signal fields, query linkage, and event aggregation from the interaction data.
What you can do after
Apply it via a generic event schema and weighted aggregation to produce query-document boosts for search.
How it connects
Connects click signals with query signals, search results, signals boosting, add-to-cart events, purchases, and relevance.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI-Powered Search. The section links point to the parts of the book it draws on.