Amazon Personalize workflow - Amazon Personalize

Amazon Personalize workflow

With Amazon Personalize, you determine your use case, import your data, train and deploy a model, and then get recommendations. Repeat the data import and training processes to sustain and improve the relevance of your recommendations as your catalogue grows. You can complete the Amazon Personalize workflow with the Amazon Personalize console, AWS Command Line Interface (AWS CLI), or the AWS SDKs.

  1. Determine your use case

    Choose your use case from the following and note its corresponding recipe type. (Recipes are Amazon Personalize algorithms prepared for different use cases.)

    • Recommending items for users (USER_PERSONALIZATION recipes)

    • Ranking items for a given user (PERSONALIZED_RANKING recipes)

    • Recommending similar items (RELATED_ITEMS recipes)

    For more information see Determining your use case.

  2. Import data

    You import item, user, and interaction records into datasets (Amazon Personalize containers for data). You can choose to import records in bulk, or incrementally, or both. With incremental imports, you can add one or more historical records or import data from real-time user activity.

    The data that you import depends on your use case. For information about the types of data that you can import, see Datasets and schemas and the sections on each dataset type (Interactions dataset, Items dataset, Users dataset).

    For more information about importing data see Preparing and importing data.

  3. Train a model

    After you've imported your data, Amazon Personalize uses it to train a model. In Amazon Personalize, you start training by creating a solution, where you specify your use case by choosing an Amazon Personalize recipe. Then you create a solution version, which is the trained model that Amazon Personalize uses to generate recommendations. For more information see Creating a solution.

  4. Deploy a model (for real-time recommendations)

    After Amazon Personalize finishes creating your solution version (trained model), you deploy it in a campaign. A campaign creates and manages a recommendation API that you use in your application to request real-time recommendations from your custom model. For more information about deploying a model see Creating a campaign. For batch recommendations, you don't need to create a campaign.

  5. Get recommendations

    Get recommendations in real time or as part of a batch workflow. Get real-time recommendations when you want to update recommendations as customers use your application. Get batch recommendations when you don't require real-time updates. For more information, see Getting recommendations.

  6. Refresh your data and repeat

    Keep your item and user data current, record new interaction data in real-time, and re-train your model on a regular basis. This allows your model to learn from your user’s most recent activity and sustains and improves the relevance of recommendations. For more information see Maintaining recommendation relevance.