[TRANSLATION ERROR] There are no messages for this key

Ensemble engine

This engine applies a mixing of many of the previous engines to make recommendations. The engine needs to be configured by setting a weight for every engine included in the setup. Then, it makes recommendations based on a weighted combination of the scores calculated by each of the individual engines.

Ensemble engine diagram

Usage

Data

The first step to make recommendations is to load items, users and interactions data into the system. It can be imported individually or in batch.

The accepted interaction types depend on the engines configured in the ensemble. All the engines in the setup must support the interaction type used to make recommendations. The supported interaction types of each engine are specified in the data section of their documentation.

Configuration

Before training, the ensemble configuration must be created with the desired engines and weights. Additional information on how to configure it can be found in Ensemble configuration.

Training

In order to make recommendations, the configured engines need to be trained. Then the model is ready to make recommendations.

You can find more details about the training in Training.

Recommendations

After the models have been trained, recommendations can be made. The user-item combinations depend on the engines configured in the ensemble. Every engine in the setup must support the recommendation type to make.

Supported interaction types:

Engineunarybinarycategoricalcontinuous
Content-based
Hybrid
Factorization Machines (FM)
ISVD
Most popular

Supported recommendations:

EngineUsers-userUsers-itemItems-itemItems-user
Content-based
Hybrid
Factorization Machines (FM)
ISVD
Most popular

Diversity

Model tags are required to apply diversification. The tags extraction is done by training the model with the Content Based engine.

Output

Every entity included in the list of recommendations comes with a percent match score. This score, a number between 0 and 100, indicates the degree of affinity between the recommended entity and the recipient.