liveBook · Manning

Examine

As explained in AI-Powered Search

An examine is the event or probability that a user consciously considers a displayed search result.

An examine is the probability that a user consciously considers a displayed search result. Unlike an impression, which only records rendering, an examine models whether the user noticed the result. This distinction helps click models separate relevance from position bias: an unexamined result should not be penalized for receiving no click. section 11.2.3

The dryer query

For a session whose last click is at rank 9, results through rank 9 count as examined, while ranks 10 and 11 do not. A document with 355 clicks among 2,707 examines receives a grade of about 0.131.

How AI-Powered Search covers it

How deep it goes
Defines examine, explains its bias rationale, and provides a session heuristic, stopping short of full click-model inference.
Examples
Uses a Nintendo result, dryer sessions, document-level click totals, and SDBN grades.
What it assumes
Assumes ranked search sessions and click logs while distinguishing conscious consideration from rendered impressions.
What you can do after
Apply it via Python to mark examined results and compute click-over-examine document grades.
How it connects
Connects impressions, clicks, click models, position bias, learning to rank, and dynamic Bayesian networks.

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.