Training method and device of transformation prediction model, electronic equipment and storage medium
By assigning weight values to sample objects and adjusting the loss function, the problem of noise influence in the training of conversion prediction models is solved, resulting in more accurate prediction of conversion behavior and improved advertising efficiency.
CN118964969BActive Publication Date: 2026-05-29TENCENT TECHNOLOGY (SHENZHEN) CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TENCENT TECHNOLOGY (SHENZHEN) CO LTD
- Filing Date
- 2023-05-15
- Publication Date
- 2026-05-29
AI Technical Summary
Technical Problem
Existing conversion prediction models suffer from overfitting during training due to low-quality and noisy historical object data, making them unable to accurately predict the probability of conversion behavior for multimedia information.
Method used
By assigning weight values to sample objects and adjusting the loss function based on probability, the initial conversion prediction model is trained to identify and reduce the impact of noise and avoid the loss of effective samples.
Benefits of technology
It improved the accuracy of conversion prediction models and the efficiency of multimedia information delivery, resulting in a 5% increase in advertisers' GMV.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN118964969B_ABST
Abstract
The application provides a training method and device of a conversion prediction model, an electronic device and a storage medium. The method comprises: predicting a probability of a first sample object performing a conversion behavior on multimedia information, the first sample object being an object that has performed a conversion behavior on the multimedia information; assigning a weight value to the first sample object based on the numerical size of the probability; adjusting a preset loss function based on the weight value of the first sample object, and training an initial conversion prediction model based on the adjusted loss function and attribute data of the first sample object to obtain a conversion prediction model for predicting a probability of a to-be-tested object performing a conversion behavior on the multimedia information. The technical solution of the embodiment of the application can improve the training efficiency of the conversion prediction model.
Need to check novelty before this filing date? Find Prior Art