Recommenders

This page explains the Product / Content Recommender trainers that are currently available on Recometry.

1. Matrix Factorization (MF)

This trainer studies and learns from the interactions between users and products. See sample training data below:

UserId	ProductId	Label
user-123	product-345	1
user-123	product-123	1
user-234	product-345	1

This agent performs best when up-selling or cross-selling to returning customers. For example:

• “Customers who bought product X also purchased products Y and Z. Since you’ve added product X to your cart, would you like to include products Y and Z as well?”
• “Users who rated product X highly also rated product Y highly. Since you’ve added product X to your cart, would you like to add product Y?”

📘

Notes:

• Avoid including any personally identifiable information (PII) or sensitive data in your training dataset. Ensure all user identifiers are properly anonymized.
• The values in the Label column must always be 1 for this trainer. This is because the dataset is designed to represent users and the products they have purchased, viewed, or interacted with.

2. Field-Aware Factorization Machine (FFM)

This trainer is optimally designed for product recommendation targeting new customers. In contrast to Matrix Factorization (MF), which exclusively models user–product interactions, this trainer incorporates categorical feature data from the training dataset to enhance recommendation performance.

UserId	Gender	ProductId	ProductCategory	Label
user-123	male	product-456	fashion	1
user-129	female	product-442	electronics	1
user-220	undisclosed	product-456	fashion	1
user-202	male	product-252	educational	1

For instance, an FFM agent trained on the specified dataset is capable of generating reliable product recommendations for a female customer identified as user-100, despite this user being absent from the training data. This capability arises from the model’s ability to generalize based on learned interactions between categorical features rather than relying solely on explicit user history.

📘

Notes:

• Avoid including any personally identifiable information (PII) or sensitive data in your training dataset. Ensure all user identifiers are properly anonymized.
• The values in the Label column must always be 1 for this trainer. This is because the dataset is designed to represent users and the products they have purchased, viewed, or interacted with.