Multi-Task Conversion Rate Training with Attribution Data Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing models for estimating conversion rates on content platforms face inefficiencies due to the imbalance in information between attribution and non-attribution data, leading to impaired accuracy and increased resource consumption when only one type of data is used for modeling.
Innovation Solution
A multi-task model training method that processes attribution and non-attribution tasks together, updating shared and independent parameters based on the results of both tasks, leveraging the larger sample size of non-attribution data to improve generalization and accuracy of the attribution task.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only attribution data is used for modeling, then the model can focus on conversion estimation, but the model accuracy is impaired due to insufficient data volume
Solution Approach 1:
The patent combines attribution data and non-attribution data into a unified multi-task model framework. The model processes both data types through shared parameters and task-specific parameters, allowing the larger non-attribution data to supplement the smaller attribution data, thereby improving overall model accuracy without requiring additional data collection.
Solution Approach 2:
The patent creates a multi-task model where a single model structure serves multiple functions: estimating conversion rates from attribution data and processing non-attribution data through shared parameters. This universal approach allows the model to leverage patterns from both data types simultaneously, improving generalization capability and estimation accuracy.
2Quantity of substance
If only non-attribution data is used for modeling, then the model can utilize larger sample size, but the model accuracy is impaired due to interference from unrelated information
Solution Approach 1:
The patent segments the model into shared parameters that process common features from both data types and task-specific parameters that handle attribution-specific information. This segmentation allows the model to process non-attribution data for general pattern learning while using task-specific parameters to focus on conversion estimation, preventing interference from unrelated information.
Solution Approach 2:
The patent applies local quality by making different parts of the model have different functions: shared parameters handle general feature extraction from both data types, while task-specific parameters handle attribution-specific conversion estimation. This allows the model to utilize the larger non-attribution data volume for general learning while maintaining focus on conversion rate estimation accuracy through specialized parameters.
3Measurement precision
If both attribution data and non-attribution data are used separately, then all information can be utilized, but resource consumption increases due to multiple modeling processes
Solution Approach 1:
The patent merges both attribution and non-attribution data processing into a single multi-task model framework with shared parameters. This unified approach allows the model to process both data types simultaneously through efficient parameter sharing, avoiding the need for separate modeling processes and reducing overall computational resource consumption while maintaining high estimation accuracy.
Data Source
AI summary
The present disclosure relates to a multi-task model training method, a data processing method, an electronic device and a storage medium. The multi-task model training method includes: obtaining training samples, where the training samples include an attribution data training sample and a non-attribution data training sample, and the training samples are constructed from conversion data corresponding to presented media content; processing the training samples through an attribution task and a non-attribution task in a multi-task model, to obtain a processing result corresponding to each task; and updating a shared parameter between the tasks in the multi-task model based on the processing result of the attribution task and the processing result of the non-attribution task, and updating an independent parameter corresponding to the attribution task based on the processing result of the attribution task.