A half-life function makes time-sensitive signals less influential as they age. In news search, recent clicks or other interactions can boost documents more strongly than older ones. section 8.5.2
The function halves a signal’s weight over each half-life interval. With a 30-day half-life, a signal receives 100% weight immediately, 75% after 15 days, 50% after 30 days, 25% after 60 days, and 12.5% after 90 days. Starting weight can vary by signal type, such as clicks, add-to-cart events, or purchases. section 8.5.2
starting_weight × 0.5(signal_age / half_life)
Computes time-adjusted signal weight by multiplying the starting weight by 0.5 raised to signal age divided by the half-life. section 8.5.2
Applying decay to query signals
A SQL aggregation can calculate time-weighted clicks, group them by query and document, and order results by the resulting boost. A 30-day half-life discounts older signals while retaining residual value across later months. section 8.5.2
