How recommendation scoring works (custom resources) - Amazon Personalize

How recommendation scoring works (custom resources)

With the User-Personalization-v2 and User-Personalization recipes, Amazon Personalize generates scores for items based on on a user's interaction data and metadata. These scores represent the relative certainty that Amazon Personalize has in whether the user will interact with the item next. Higher scores represent greater certainty.


Amazon Personalize doesn't show scores for domain recommenders or the Similar-Items, SIMS or Popularity-Count recipes. For information on scores for Personalized-Ranking recommendations, see How personalized ranking scoring works.

Amazon Personalize generates scores for items relative to each other on a scale from 0 to 1 (both inclusive). With User-Personalization-v2, Amazon Personalize generates scores for a subset of your items. With User-Personalization, Amazon Personalize scores all of the items in your catalog.

If you use User-Personalization-v2 and apply a filter to recommendations, depending on how many recommendations the filter removes, Amazon Personalize might add placeholder items. It does this to meet the numResults for your recommendation request. These items are popular items, based on amount of interactions data, that satisfy your filter criteria. They don't have a relevance score for the user.

For both User-Personalization-v2 and User-Personalization, the total of all scores equals 1. For example, if you're getting movie recommendations for a user and there are three movies appearing the Items dataset and Interactions dataset, their scores might be 0.6, 0.3, and 0.1. Similarly, if you have 10,000 movies in your inventory, the highest-scoring movies might have very small scores (the average score would be.001), but, because scoring is relative, the recommendations are still valid.

In mathematical terms, scores for each user-item pair (u,i) are computed according to the following formula, where exp is the exponential function, w̅u and wi/j are user and item embeddings respectively, and the Greek letter sigma (Σ) represents summation over all items with scores:

Depicts the formula used to calculate scores for each item in recommendations.