Maintaining recommendation relevance - Amazon Personalize
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Maintaining recommendation relevance

Relevant recommendations can 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 and custom resources up to date. This allows Amazon Personalize to learn from your user’s most recent behavior and include your newest items in recommendations.

Keeping datasets current

As your catalog grows, update your historical data with bulk or individual data import operations. For more information about importing historical data, see Step 2: Preparing and importing data. For information on how data you import after training a model influences recommendations, see How new data influences real-time recommendations.

For use cases and recipes that provide personalized real-time recommendations, keep your Item interactions dataset up to date with your users' behavior. Do this by recording item interactions with an event tracker and the PutEvents API operation. Amazon Personalize updates recommendations based on your user's most recent activity as they interact with your catalog. For information about real-time personalization, see Real-time personalization. For more information on recording real-time events, see Recording events.

Maintaining domain recommenders

Amazon Personalize automatically retrains the models backing your recommenders every 7 days. This is a full retraining that creates entirely new models based on the entirety of the data in your datasets. If you modify the columns used in training, Amazon Personalize automatically starts a full retraining of the models backing your recommender.

  • For Top picks for you and Recommended for you use cases, Amazon Personalize updates your recommender to consider new items for recommendations. Automatic updates are not a full retraining where the model learns from your users' behavior. Instead, automatic updates allow Amazon Personalize to feature your new items in recommendations before the recommender's next full retraining. For information about automatic updates, see Automatic updates.

  • If you use the Trending now use case, Amazon Personalize automatically evaluates your interactions data every two hours and identifies trending items. You don't have to wait for your recommender to retrain.

While recommender retraining is in progress, you can still get recommendations from the recommender. Until the retraining completes, the recommender uses the previous configuration and models. To track updates, you can view the timestamp for the latest recommender update on the Recommender details page in the Amazon Personalize console. Or you can view the latestRecommenderUpdate details from the DescribeRecommender operation.

Maintaining custom solutions

Maintain your custom solutions by retraining regularly. Create a new solution version (retrain the model) to include new items in recommendations and update the model with your user’s most recent behavior.

Your retraining frequency depends on your business requirements and the recipe that you use. For all recipes, we recommend creating a new solution version at least weekly. This creates a completely new model based on the entirety of the training data from the datasets in your dataset group. For User-Personalization, you must set trainingMode to FULL for a full retraining.

If you add new items frequently, you might need to retrain more frequently depending on your recipe:

  • If you don't use a recipe with automatic updates (such as User-Personalization or Next-Best-Action) or the Trending-Now recipe, you must create a new solution version for Amazon Personalize to consider the new items for recommendations.

  • If you use User-Personalization or Next-Best-Action, Amazon Personalize automatically updates your latest fully trained solution version to consider new items for recommendations.

    Automatic updates are not a full retraining where the model learns from your users' behavior. Instead, automatic updates allow Amazon Personalize to feature your new items in recommendations before the next full retraining.

    You should still train a new solution version weekly with trainingMode set to FULL. If every two hours is not frequent enough, you can manually create a solution version with trainingMode set to UPDATE to consider those new items for recommendations. Just remember that Amazon Personalize automatically updates only your latest fully trained solution version. The manually updated solution version won't be automatically updated in the future.

    For more formation about auto updates, including additional guidelines and requirements, see Automatic updates.

  • If you use Trending-Now, Amazon Personalize automatically identifies the top trending items in your interactions data over a configurable interval of time. You don't have to manually create a new solution version for Trending-Now to consider new items from bulk or incremental interactions since the last training. For more information, see Trending-Now recipe.

For information on creating a new solution version, see Creating a solution version. After you create a new solution version, you must update the campaign to deploy it. For more information, see Updating a campaign.

You can automate and schedule re-training and data import tasks with Maintaining Personalized Experiences with Machine Learning, an Amazon Solutions Implementation that automates the Amazon Personalize workflow, including data import, solution version training, and batch workflows. For more information see Maintaining Personalized Experiences with Machine Learning.