This article describes the main concepts and terminology behind the Amazon Personalize service.


Contents


Personalize Datasets

Input data from Redshift must be mapped to the Personalize standard dataset so that it can be used to train your model.

The service uses three types of datasets:


Interactions

This dataset is always required. 

It stores historical and real-time data from interactions between users and items. This could include contextual data such as impressions, clicks, user reviews, or ratings.


Users

This dataset is optional. 

It stores metadata about users and could include information such as age, gender or loyalty membership.


Items

This dataset is optional. 

It stores metadata about items and could include information such as price, SKU type or availability. 


Personalize Recipes (models)

Recipes are Amazon Personalize algorithms that are available for specific use cases.

Hyperparameters for all recipes can be fine-tuned to improve the recommendations from your trained models.

The following four recipes are available:


User Personalization

The User Personalization recipe predicts the items that a user will interact with based on Interactions, Items, and User datasets. 


Popularity Count

The Popularity Count recipe recommends the most popular items based on all of your user behavioural data. The most popular items have the most interactions with unique users. The recipe returns the same popular items for all users and provides a good baseline for comparing with other recipes.


Personalized Ranking

The Personalized Ranking recipe generates personalized rankings of items. A personalized ranking is a list of recommended items that are re-ranked for a specific user. This is useful if you have a collection of ordered items, such as search results, promotions, or curated lists, and you want to provide a personalized re-ranking for each of your users.


Similar Items

 The Item-to-item similarities (SIMS) recipe uses collaborative filtering to recommend items that are most similar to an item you specify when you get recommendations. SIMS uses the Interactions dataset, not item metadata such as colour or price, to determine similarity. SIMS identifies the co-occurrence of the item in user histories in your Interaction dataset to recommend similar items. For example, with SIMS Amazon Personalize could recommend coffee shop items customers frequently bought together or movies that different users also watched.

Training is faster with the SIMS recipe compared to other recipes. If there isn't sufficient user behaviour data for an item or the item ID you provide isn't found, SIMS recommends popular items.


Solutions

A trained model is referred to as a solution.

This is the trained model deployed on top of your selected recipe, using your custom / hyperparameter optimization tuned parameters.

All solutions are versioned; after every new model training process, a new version of the solution is created. 

You can track and monitor your Amazon Personalize experiments from Peak’s Model Management function. After your Personalize workflow has run successfully, the system creates a listing in the Model management Experiments screen.


Campaigns

A campaign is a deployment that Amazon Personalize creates on top of any existing solution version, to enable real-time recommendations to be served.

All campaigns can scale up and down automatically, but it is still recommended to pre-assess your requirements and configure with the correct scaling parameters.