Getting batch recommendations and user segments
To get recommendations and user segments for datasets that do not require real-time updates, you can use an asynchronous batch workflow with historical data only. For example, you might get product recommendations for all users on an email list or item-to-item similarities across an inventory. Or with the USER_SEGMENTATION recipes, get user segments for data-driven advertising based on items in your inventory and your user's interactions. You don't need to create an Amazon Personalize campaign to get batch recommendations or user segments.
-
To get batch recommendations, you use a batch inference job. A batch inference job is a tool that imports your batch input data from an Amazon S3 bucket, uses your solution version to generate item recommendations, and then exports the recommendations to an Amazon S3 bucket.
Depending on the recipe, the input data can be a list of users or items or list of users each with a collection of items. Your input data must be in JSON format. Use a batch inference job when you want to get batch item recommendations for your users or find similar items across an inventory.
-
To get user segments, you use a batch segment job. A batch segment job is a tool that imports your batch input data from an Amazon S3 bucket, uses your solution version trained with a USER_SEGMENTATION recipe to generate user segments for each row of input data, and exports the segments to an Amazon S3 bucket. Each user segment is sorted in descending order based on the probability that each user will interact with items in your inventory. When generating user segments, Amazon Personalize considers all data that you imported individually (including streamed interactions after you create a new solution version), but only bulk data imported that you imported with an import mode of FULL (replacing existing data).
Depending on the recipe, the input data is a list of items or item metadata attributes in JSON format. For item attributes, your input data can include expressions to create user segments based on multiple metadata attributes. Use a batch segment job when you use a USER_SEGMENTATION recipe and want to get segments of users who will most likely interact with each of your items in your inventory.
For both batch workflows, we recommend that you use a different location for your output data (either a folder or a different Amazon S3 bucket).
You can then run multiple batch inference jobs using the same input data location.
To use data you record in real time with the PutEvents API operation,
you must retrain your solution version before creating a batch inference job or batch segment job. If you use a filter with placeholder parameters, you must include the values for the parameters
in your input data in a filterValues object. For more information, see Providing filter values in your input JSON.
The batch workflow is as follows:
-
Prepare and upload your input data in JSON format to an Amazon S3 bucket. The format of your input data depends on the recipe you use and the job you are creating. See Preparing and importing batch input data.
-
Create a separate location for your output data, either a folder or a different Amazon S3 bucket.
-
Create a batch inference job or a batch segment job. See Creating a batch inference job (console), Creating a batch inference job (AWS CLI), or Creating a batch inference job (AWS SDKs).
-
When the batch inference or batch segment job is complete, retrieve the recommendations or user segments from your output location in Amazon S3.
Topics
Batch workflow permissions requirements
For batch workflows, your Amazon Personalize IAM service role needs permission to access and add files to your Amazon S3 buckets. For information on granting permissions, see Service-linked role policy for batch workflows. For more information on bucket permissions, see User policy examples in the Amazon Simple Storage Service Developer Guide.
Amazon S3 buckets and objects must be either encryption free or, if you are using AWS Key Management Service (AWS KMS) for encryption, you must grant Amazon Personalize and your Amazon Personalize IAM service role permission to use your key. For more information, see Giving Amazon Personalize permission to use your AWS KMS key.
Batch workflow scoring
With User-Personalization and Personalized-Ranking recipes, Amazon Personalize calculates batch inference job recommendation scores as described in How User-Personalization recommendation scoring works and How personalized ranking scoring works. You can view scores in the batch inference job's output JSON file. Scores are only returned by models trained with the User-Personalization and Personalized-Ranking recipes.
