A training method of a prediction model, a content promotion method, and related devices
By matching the model structure and evaluating hyperparameter combinations based on historical data from content promoters, the problem of low prediction model accuracy in existing technologies is solved, achieving higher prediction accuracy and automated hyperparameter selection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TENCENT TECHNOLOGY (SHENZHEN) CO LTD
- Filing Date
- 2024-12-03
- Publication Date
- 2026-06-05
AI Technical Summary
In existing technologies, the prediction accuracy of the target prediction model trained is low, mainly due to the influence of factors such as sample comprehensiveness, model structure and learning rate, and the low adaptability of the prediction model caused by the differences in content characteristics of different content providers.
By acquiring historical promotion data from content promoters, a model structure matching their promotion positioning information is determined. Through multiple rounds of iterative training and evaluation of various hyperparameter value combinations, the hyperparameter combination that meets the training objectives is selected to form a target prediction model.
It improves the prediction accuracy of the prediction model, avoids the instability of prediction accuracy caused by factors such as sample comprehensiveness, model structure or learning rate, and enhances the automation and quantitative selection of hyperparameter settings.
Smart Images

Figure CN122155786A_ABST