liveBook · Manning

Signal age

As explained in AI-Powered Search

Signal age is the elapsed time since an interaction signal occurred.

Signal age is the elapsed time since an interaction signal occurred. Time-decay weighting reduces older signals’ influence while preserving residual value for longer-lived documents. section 8.5.2

starting_weight × 0.5(signal_age / half_life)

Calculate a signal’s time-weighted contribution as starting weight × 0.5^(signal age ÷ half-life). section 8.5.2

A 30-day half-life assigns approximately:

  • 100% weight to a current signal, 75% to a 15-day-old signal, and 50% to a 30-day-old signal. section 8.5.2
  • 25% weight to a 60-day-old signal and 12.5% to a 90-day-old signal. section 8.5.2

Applying signal age in aggregation

A signals query can group the latest click for each user, query, and document, apply a 30-day half-life with POW(0.5, signal_age / 30), sum boosts by query and document, and order results by boost. section 8.5.2

How AI-Powered Search covers it

How deep it goes
Explains the decay calculation, configurable half-life and weights, and a SQL aggregation implementation.
Examples
Uses half-life weight sequences and a SQL query that aggregates time-weighted click signals by query and document.
What it assumes
Assumes signal timestamps, signal types, query-document relationships, and a target date for calculating age.
What you can do after
Apply time decay through configurable half-lives and signal weights in a signals aggregation query.
How it connects
Connects signal age with signals boosting, half-life, decay functions, signal types, and ranking.

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.