Posted On: Jul 2, 2020
Amazon Personalize uses machine learning to personalize recommendations for products, content, and marketing communications for your users, with no machine learning experience required. This technology has been perfected from over 20 years of recommender systems development at Amazon.com.
Today, we are pleased to announce improved handling of missing or sparse metadata for interactions, user and item dataset types, in Amazon Personalize. Metadata such as brand of a product, age group of a user, or device type for the browsing session can be useful for improving the accuracy and relevance of recommendation models. However, this data is often imperfect, and can have data missing in many cases, which if not handled carefully during a machine learning model training process, can adversely affect the model performance.
To handle such situations, Amazon Personalize now enables you to define “null” as an acceptable value in a schema when creating an Amazon Personalize solution. This ensures that imperfect metadata can safely be used for improving the relevance of your recommendations. To use this feature, you can define “null” as an allowed metadata value in the Amazon Personalize console or API when defining the schema for your dataset. Next, you can safely import your dataset into Amazon Personalize and create a solution. While creating a solution, Amazon Personalize will automatically recognize the fields with missing metadata and handle them appropriately while training the machine learning model. To learn more about this feature, visit our developer guide.
Improved Handling of Missing Metadata in Amazon Personalize is now available in US East (N. Virginia, Ohio), US West (Oregon), Canada (Central), Europe (Ireland),and Asia Pacific (Sydney, Tokyo, Mumbai, Singapore, Seoul).