Maintaining recommendation relevance (Domain dataset group) - Amazon Personalize

Maintaining recommendation relevance (Domain dataset group)

Maintain the relevance of recommendations to increase user engagement, click-through rate, and conversion rate for your application as your catalogue grows. To maintain and improve the relevance of Amazon Personalize recommendations for your users, keep your data in Amazon Personalize up to date. This allows Amazon Personalize to learn from your user’s most recent behavior and include your newest items in recommendations.

For users and items, as your catalog grows, update your historical data with bulk or individual data import operations. We recommend that you first import your records in bulk, and then add individual items and users as your catalog grows. For information on managing item and user data, see Managing data.

For interactions data, keep your Interactions dataset up to date with your users' behavior by recording interaction events with the PutEvents operation. Amazon Personalize updates recommendations based on your user's most recent activity as they interact with your application. For more information on recording real-time events, see Recording events.

New records influence recommendations as follows:

  • For new events, Amazon Personalize immediately uses real-time interaction events between a user and existing items (items you included in the data you used to train the latest model) when generating recommendations for the same user. For more information, see How real-time events influence recommendations.

  • For new items, if you create the recommender with Top picks for you and Recommended for you, Amazon Personalize automatically updates the underlying models every two hours. After each update, the new items might be included in recommendations.

    For any other domain use case, Amazon Personalize automatically trains new models for your recommenders every 7 days, starting from the recommender creation date.

  • For new users without interactions data, recommendations are initially for only popular items. To get relevant recommendations for the user, you can import bulk interactions data for the user and wait for the next model update. Or you can record events in real-time for the user as they interact with items. Their recommendations will be more relevant as you record more events. For more information, see Recording events.