Training method of recommendation model, recommendation method, device and related equipment
By acquiring sample data from multiple sub-confirmation results, distinguishing between positive, negative, and unconfirmed samples, and training a multi-task recommendation model using a feature representation module and a gating network module, the problem of insufficient utilization of confirmation results in existing technologies is solved, thereby improving the model's accuracy and advertising effectiveness.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUAWEI TECH CO LTD
- Filing Date
- 2025-01-17
- Publication Date
- 2026-07-17
AI Technical Summary
Existing recommendation models fail to fully utilize the information in the confirmation results during training, resulting in low accuracy. This is especially true in multi-task recommendation models, where the confusion between unconfirmed samples and negative samples leads to insufficient prediction accuracy.
By acquiring sample data including multiple sub-confirmation results, the sample labels of the sample data are determined. A multi-task recommendation model is used to distinguish positive samples, negative samples, and unconfirmed samples. A feature representation module and a gating network module are used to control the sample data to participate in training, thereby improving the accuracy of the model.
It improves the training effect of multi-task recommendation models, enhances the ability to identify unconfirmed samples, and improves the prediction accuracy and advertising performance of the models.
Smart Images
Figure CN122414367A_ABST