Learn how and why we use the weighting curve on product recommendation data.
Understanding the weighting curve
By default, product recommendations look at data over a period of all time. To handle situations where very old data skews results and less relevant products are recommended we apply a weighting curve.
The curve’s formula considers data across a timeline, from the date of the first event (such as an order or page view) to the most recent event. Results are then scored between 0.1 and 1 based on their position in the timeline.
The weighting curve strongly favours more recent events. Only items placed in the last 20% of the timeline receive a weighted score above 0.5. From there, the weighting rises rapidly to 1.
Custom weighting curves
The products you sell, customer type, seasonality, and other factors may affect how you think about the recency of your order or Web Behavior Tracking data.
If you would to like try a custom weighting curve, please speak to your account manager.