Long-Tail Boost MF

Weighted Regularized Matrix Factorization (WRMF) on implicit feedback, with a long-tail reweighting term that down-weights frequently exposed items so latent collaborative signal is combined with catalog diversity.

Algorithm

Config value: "approach": "behaviorAlgos", "sub_approach": "longTailBoostMF"

The learner fits latent factors for users and items using Alternating Least Squares (ALS) on implicit feedback matrices. The final score blends the dot product of user and item vectors with an inverse-exposure factor:

\(\text{score} = (\mathbf{u} \cdot \mathbf{v}) \cdot \frac{1}{(\text{exposure} + 1)^{\gamma}} \cdot \text{variableMultiplier}\)

Here \(\mathbf{u}\) and \(\mathbf{v}\) are the user and item latent vectors, \(\gamma\) controls how strongly popularity is penalized, and variableMultiplier carries any configured scaling from the deployment.

Parameters

• latentDim (\(k\)): Latent factor dimensionality. Default: 6.
• numIters: ALS iterations. Default: 8.
• regLambda (\(\lambda_{\text{reg}}\)): L2 regularization for factors. Default: 0.05.
• gamma (\(\gamma\)): Long-tail / inverse-exposure exponent. Default: 0.4.
• entropyW: Optional entropy-related weighting term (default 0.0 when unused).
• Processing Window: Time window in milliseconds for historical data.
• Historical Count: Max records to process per update cycle.

Cold Start

Recommendations are always returned. The real-time training path in RollingBehavior always produces a scored options array:

• No history: Every offer in the options store receives a uniform random score. All offers are ranked and passed to the post-score class.
• Sparse history: The model initializes unseen users and items with small Gaussian noise (\(\mathcal{N}(0, 0.01^2)\)), so algorithm-level scores cluster near zero. Unscored offers receive a random fallback score.
• Sufficient history: The ALS-learned latent factors produce meaningful collaborative-filtering scores, amplified by the long-tail reweighting.

The scored options are then sorted by arm_reward and handed to the configured dynamic post-score class, which controls the final offer selection and response formatting.

When To Use

• Collaborative filtering with many user–item events and a desire to promote the long tail
• Catalogs where popularity bias should be counteracted after you already have interaction volume

When NOT To Use

• Cold start deployments or very sparse interaction logs
• Few interactions per user or item (use simpler bandit methods first)
• Simple recommendation problems where matrix factorization is unnecessary

Example

from prediction.apis import deployment_management as dm
from prediction.apis import online_learning_management as ol
from prediction import jwt_access
 
auth = jwt_access.Authenticate("http://localhost:3001/api", ecosystem_username, ecosystem_password)
 
deployment_id = "demo-long-tail-boost-mf"
 
online_learning_uuid = ol.create_online_learning(
        auth,
        algorithm="ecosystem_rewards",
        name=deployment_id,
        description="Long-Tail Boost MF configuration",
        feature_store_collection="set_up_features",
        feature_store_database="my_mongo_database",
        options_store_database="my_mongo_database",
        options_store_collection="demo-deployment_options",
        randomisation_processing_count=5000,
        randomisation_processing_window=604800000,
        contextual_variables_offer_key="offer",
        create_options_index=True,
        create_covering_index=True
)
 
online_learning = dm.define_deployment_multi_armed_bandit(epsilon=0, dynamic_interaction_uuid=online_learning_uuid)
 
parameter_access = dm.define_deployment_parameter_access(
    auth,
    lookup_key="customer_id",
    lookup_type="string",
    database="my_mongo_database",
    table_collection="customer_feature_store",
    datasource="mongodb"
)
 
deployment_step = dm.create_deployment(
    auth,
    project_id="demo-project",
    deployment_id=deployment_id,
    description="Long-Tail Boost MF demo deployment",
    version="001",
    plugin_post_score_class="PlatformDynamicEngagement.java",
    plugin_pre_score_class="PreScoreDynamic.java",
    scoring_engine_path_dev="http://localhost:8091",
    mongo_connect=f"mongodb://{mongo_user}:{mongo_password}@localhost:54445/?authSource=admin",
    parameter_access=parameter_access,
    multi_armed_bandit=online_learning
)