Creating recommenders - Amazon Personalize
Minimum recommendation requests per second and auto-scalingEnabling metadata in recommendationsConfiguring columns used when training

Creating recommenders

After you create a Domain dataset group and import data, you can create recommenders for your domain use cases. A recommender is a Domain dataset group resource that generates recommendations. You use a recommender in your application to get real-time recommendations with the GetRecommendations operation.

When you create a recommender, you specify a use case and Amazon Personalize trains the models backing the recommender with the best configurations for the use case. Each use case has different API requirements for getting recommendations. For a list of recommender use cases by domain, see Choosing a use case. You can create at most 15 recommenders per region.

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. With Top picks for you and Recommended for you use cases, Amazon Personalize updates the existing models every two hours to include new items in recommendations with exploration.

When you create a recommender, you can enable item metadata in recommendations. For more information, see Enabling metadata in recommendations.

You can create recommenders with the Amazon Personalize console, AWS Command Line Interface (AWS CLI), or AWS SDKs.

Recommender statuses

A recommender can be in one of the following states:

  • CREATE PENDING > CREATE IN_PROGRESS > ACTIVE -or- CREATE FAILED

  • DELETE PENDING > DELETE IN_PROGRESS

To get the recommender status, navigate to the Recommenders page in the Amazon Personalize console or use the DescribeRecommender operation.

Topics

Minimum recommendation requests per second and auto-scaling

Important

A high minRecommendationRequestsPerSecond will increase your bill. We recommend starting with 1 for minRecommendationRequestsPerSecond (the default). Track your usage using Amazon CloudWatch metrics, and increase the minRecommendationRequestsPerSecond as necessary.

When you create a recommender, you can configure the recommender's minimum recommendation requests per second. The minimum recommendation requests per second (minRecommendationRequestsPerSecond) specifies the baseline recommendation request throughput provisioned by Amazon Personalize. The default minRecommendationRequestsPerSecond is 1. A recommendation request is a single GetRecommendations operation. Request throughput is measured in requests per second and Amazon Personalize uses your requests per second to derive your requests per hour and the price of your recommender usage.

If your requests per second increases beyond minRecommendationRequestsPerSecond, Amazon Personalize auto-scales the provisioned capacity up and down, but never below minRecommendationRequestsPerSecond. There's a short time delay while the capacity is increased that might cause loss of requests.

Your bill is the greater of either the minimum requests per hour (based on minRecommendationRequestsPerSecond) or the actual number of requests. The actual request throughput used is calculated as the average requests/second within a one-hour window. We recommend starting with the default minRecommendationRequestsPerSecond, track your usage using Amazon CloudWatch metrics, and then increase the minRecommendationRequestsPerSecond as necessary.

Enabling metadata in recommendations

Important

When you enable metadata in recommendations, you incur additional costs. For more information see Amazon Personalize pricing.

When you create a recommender, you can enable the option to include item metadata from your Items dataset with recommendation results. If enabled, you can specify the columns from your Items dataset in your request for recommendations. Amazon Personalize returns this data for each item in the recommendation response.

You might use metadata to enrich recommendations in your user interface, such as adding the genres for movies to carousels. Or you might use it to visually assess recommendation quality. If you use generative AI in your app, you can plug the metadata into AI prompts to generate more relevant content. For more information about using Amazon Personalize with generative AI, see Amazon Personalize and generative AI.

To add metadata to recommendations, you must have an Items dataset with a column of metadata. You don't have to use the metadata in training. For information about creating a dataset, see Creating a dataset and a schema. For information managing and updating data, see Managing the training data in your datasets.

Configuring columns used when training

When you create a recommender, you can modify the columns Amazon Personalize considers when training the models backing your recommender.

You might do this to experiment with different combinations of training data. Or you might exclude columns without meaningful data. For example, you might have a column that you want to use only to filter recommendations. You can exclude this column from training and Amazon Personalize considers it only when filtering.

You can't exclude EVENT_TYPE columns. By default, Amazon Personalize uses all columns that can be used when training. The following data is always excluded from training:

  • Columns with the boolean data type

  • Impressions data

  • Custom string fields that aren't categorical or textual

You can't include impressions data in training, but if your use case or recipe uses it, Amazon Personalize uses impressions data to guide exploration when you get recommendations.