Amazon Personalize terms

The following terms relate to importing, exporting, and formatting data in Amazon Personalize.

actions dataset

A container for metadata about your actions. An action is an engagement or revenue generating activity that you might want to recommend to your users, such as installing your mobile app, or joining your loyalty program. Metadata for actions might include the action's expiration timestamp, value, repeat frequency data, and categorical metadata. This type of data is used only by the Next-Best-Action recipe.

actions interactions dataset

A container for historical and real-time data that you collect from interactions between users and actions. Each action interaction consists of a userID, actionID, timestamp, event type, and any additional data about the interaction, such as categorical metadata. This type of data is used only by the Next-Best-Action recipe.

contextual metadata

Interactions data that you collect about a user's browsing context (such as device used or location) when an event (such as a click) occurs. Contextual metadata can improve recommendation relevance for new and existing users.

dataset

A container for data that you upload to Amazon Personalize. There are five types of Amazon Personalize datasets: Users, Items, Item interactions dataset, and Actions.

dataset group

A container for Amazon Personalize resources, including datasets, domain recommenders, and custom resources. A dataset group organizes your resources into independent collections, where resources from one dataset group can't influence resources in any other dataset group. A dataset group can either be a Domain dataset group or a Custom dataset group.

Domain dataset group

A dataset group containing preconfigured resources for different business domains and use cases. Amazon Personalize manages the life cycle of training models and deployment. When you create a Domain dataset group, you choose your business domain, import your data, and create recommenders for each of your use cases. You use your recommender in your application to get recommendations with the GetRecommendations operation.

If you start with a Domain dataset group, you can still add custom resources such as solutions and solution versions trained with recipes for custom use cases.

Custom dataset group

A dataset group containing only custom resources, including solutions, solution versions, filters, campaigns, and batch inference jobs. You use a campaign to get recommendations with the GetRecommendations operation. You manage the life cycle of training models and deployment. If you start with a Custom dataset group, you can't associate it with a domain later. Instead, create a new Domain dataset group.

dataset export job

A record export tool that outputs the records in a dataset to one or more CSV files in an Amazon S3 bucket. The output CSV file includes a header row with column names that match the fields in the dataset's schema.

dataset import job

A bulk import tool that populates your Amazon Personalize dataset with data from a CSV file in your Amazon S3 bucket.

event

A user action – such as a click, a purchase, or a video viewing – that you record and upload to an Amazon Personalize Item interactions dataset. You import events in bulk from a CSV file, incrementally with the Amazon Personalize console, and in real-time.

explicit impressions

A list of items that you manually add to an Amazon Personalize Item interactions dataset. Unlike implicit impressions, which Amazon Personalize automatically derives from your recommendation data, you choose what to include in explicit impressions.

implicit impressions

The recommendations that your application shows a user. Unlike explicit impressions, which you manually add to an Item interactions dataset, Amazon Personalize automatically derives implicit impressions from your recommendation data.

impressions data

The list of items that you presented to a user when they interacted with a particular item by clicking it, watching it, purchasing it, and so on. Amazon Personalize uses impressions data to calculate the relevance of new items for a user based on how frequently users have selected or ignored the same item.

interactions dataset

A container for historical and real-time data that you collect from interactions between users and items (called events). Interactions data can include event type data and contextual metadata.

items dataset

A container for metadata about your items, such as price, genre, or availability.

repeat frequency

A type of action metadata you can import into an Actions dataset. Repeat frequency data specifies how many days Amazon Personalize should wait to recommend a particular action after a user interacts with it, based on the user's history in your Action interactions dataset.

schema

A JSON object in Apache Avro format that tells Amazon Personalize about the structure of your data. Amazon Personalize uses your schema to parse your data.

users dataset

A container for metadata about your users, such as age, gender, or loyalty membership.

The following terms relate to training a model in Amazon Personalize.

item-to-item similarities (SIMS) recipe: A RELATED_ITEMS recipe that uses the data from an Interactions dataset to make recommendations for items that are similar to a specified item. The SIMS recipe calculates similarity based on the way users interact with items instead of matching item metadata, such as price or color.
item-affinity: A USER_SEGMENTATION recipe that uses the data from an Item interactions dataset and Items dataset to create user segments for each item that you specify based on the likelihood that the users will interact with the item.
item-attribute-affinity: A USER_SEGMENTATION recipe that uses the data from an Item interactions dataset and Items dataset to create a user segment for each item attribute that you specify based on the likelihood that the users will interact with items with the attribute.
Next-Best-Action recipe: This recipe generates real-time recommendations for the next best actions for your users. The next best action for a user is the action that they will most likely take. For example, enrolling in your loyalty program, downloading your app, or applying for a credit card. For more information see Next-Best-Action recipe.
Personalized-Ranking-v2 recipe: A PERSONALIZED_RANKING recipe that ranks a collection of items that you provide based on the predicted interest level for a specific user. This recipe uses a transformer based architecture to train a model that learns from item interactions data, item metadata and user metadata. Use the Personalized-Ranking-v2 recipe to personalize the order of curated lists of items or search results that are personalized for a specific user. It can train on up to 5 million items and generate more relevant recommendations with lower latency than the previous version.
personalized-ranking recipe: A PERSONALIZED_RANKING recipe that ranks a collection of items that you provide based on the predicted interest level for a specific user. Use the personalized-ranking recipe to personalize the order of curated lists of items or search results that are personalized for a specific user.
popularity-count recipe: A USER_PERSONALIZATION recipe that recommends the items that have the most interactions with unique users.
recommender: A Domain dataset group tool that generates recommendations. You create a recommender for a Domain dataset group and use in your application to get real-time recommendations with the GetRecommendations API. 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.
recipe: An Amazon Personalize algorithm that is preconfigured to predict the items that a user will interact with (for USER_PERSONALIZATION recipes), or calculate items that are similar to specific items that a user has shown interest in (for RELATED_ITEMS recipes), or rank a collection of items that you provide based on the predicted interest for a specific user (for PERSONALIZED_RANKING recipes).
solution: The recipe, customized parameters, and trained models (Solution Versions) that Amazon Personalize uses to generate recommendations.
solution version: A trained model that you create as part of a solution in Amazon Personalize. You deploy a solution version in a campaign to activate the personalization API that you use to request recommendations.
training mode: The scope of training to be performed when creating a solution version. There are two different modes: FULL and UPDATE. FULL mode creates a completely new solution version based on the entirety of the training data from the datasets in your dataset group. UPDATE incrementally updates the existing solution version to recommend new items that you added since the last training.

Note
With User-Personalization-v2, User-Personalization, or Next-Best-Action, Amazon Personalize automatically updates the latest solution version trained with FULL training mode. See Automatic updates.
User-Personalization-v2 recipe: A USER_PERSONALIZATION recipe that recommends items a user will interact with based on their preferences. This recipe uses a transformer based architecture to train a model that learns from item interactions data, item metadata, and user metadata. It can train on up to 5 million items and generate more relevant recommendations with lower latency than the previous version.
User-Personalization recipe: A Hierarchical Recurrent Neural Network (HRNN) based USER_PERSONALIZATION recipe that predicts the items that a user will interact with. The user-personalization recipe can use item exploration and impressions data to generate recommendations for new items.

The following terms relate to deploying and using a model to generate recommendations.

action optimization period: The period of time Amazon Personalize uses when predicting the actions that the user will most likely take. For example, if the action optimization period is 14 days, Amazon Personalize predicts the actions users will most likely take in the next 14 days. You configure the action optimization period when you create a solution with the Next-Best-Action recipe.
batch inference job: A tool that imports your batch input data from an Amazon S3 bucket, uses your solution version to generate recommendations, and exports the recommendations to an Amazon S3 bucket. We recommend using a different location for your output data (either a folder or a different Amazon S3 bucket). Use a batch inference job to get recommendations for large datasets that do not require real-time updates.
batch segment job: A tool that imports your batch input data from an Amazon S3 bucket, uses your solution version to create user segments, and exports the user segments to an Amazon S3 bucket. We recommend using a different location for your output data (either a folder or a different Amazon S3 bucket). Use a batch segment job with a solution backed by a USER_SEGMENTATION recipe to create segments of users based on the likelihood the user will interact with different items or items with different item attributes.
campaign: A deployed solution version (trained model) with provisioned dedicated transaction capacity for creating real-time recommendations for your application users. After you create a campaign, you use the getRecommendations or getPersonalizedRanking API operations to get recommendations.
item exploration: With exploration, recommendations include some items or actions that would be typically less likely to be recommended for the user, such as new items or actions, items or actions with few interactions, or items or actions less relevant for the user based on their previous behavior.
metric attribution: A tool you use to measure the impact of item recommendations. A metric attribution creates reports based on the item interactions and items data that you import, and the metrics that you specify. For example, the total length of movies watched by users, or the total number of click events.
recommendations: A list of items that Amazon Personalize predicts a user will interact with. Depending on the Amazon Personalize recipe used, recommendations can be either a list of items (USER_PERSONALIZATION recipes and RELATED_ITEMS recipes), or a ranking of a collection of items you provided (PERSONALIZED_RANKING recipes).
user segments: Lists of user that Amazon Personalize predicts a user will interact with your catalogue. Depending on the USER_SEGMENTATION recipe used, you create user segments based on items (Item-Affinity recipe) item metadata (Item-Attribute-Affinity recipe). You create user segments with a batch segment job.

Topics

Data import and management

Training

Note

Model deployment and recommendations