Overview

The Collaborative Filtering Forecast Processor calculates product recommendations for users, using the existing user base and a selection of products that are already chosen/rated based on the ALS (Alternating Least Squares) algorithm.

This processor is currently in an experimental state.

Note: Changed machine learning algorithm implementations in Spark 3 may slightly change results compared to Spark 2.

Input

This processor has two input nodes:

The left input node needs to be connected to the training dataset which has to contain at least three columns:

• User ID: column identifying users (must be of type "Int")
• Product ID: column identifying products (must be of type "Int")
• Ratings/Purchases: column containing purchases/ratings of the products by the user base (must be of type "Int", "Double" or "Numeric")

The right input node needs to be connected to the forecast dataset that should contain information about the products that were already purchased/rated. It includes at least two columns: the Product ID and Ratings/Purchases column.

NOTE THAT: The Ratings/Purchases columns in training and forecast datasets must have the exact same name and type.

Configuration

NOTE THAT: The amount of recommendations (fifth field) can only vary from one to twenty, and the actual value may be less than the amount entered in case the suitable recommendations are not enough.

Output

The Collaborative Filtering Forecast Processor generates a table containing the recommended products and the expected purchases/ratings of the new user.

Example

In the following example two Custom Input Table processors will be used to generate training and forecast datasets that will be fed to the Collaborative Filtering Forecast Processor.

Example Input

Training Data

Products of the new user

Workflow

Example Configuration

Result

Related Articles

ALS Recommender Processor

Association Rule Generation Processor

Association Rule Application Processor