Recommendation Model Training via Sample Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In machine learning modeling, samples obtained through manual annotation and user behavior extraction often contain noise due to misoperation and cheating, which significantly impacts the training of recommendation models.
Innovation Solution
A method is provided for training a model in an information recommendation system, where a reference training sample is purified and denoised using a sample model, and an updated target model is obtained by inputting the purified sample into a target model, enabling the system to replace manual rules with a machine learning scheme and improve the model's accuracy, generalization, and transfer ability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If samples are obtained from manual annotation and user behavior extraction, then the model training process can be completed, but the samples contain a lot of noise which significantly impacts the model training effect
Solution Approach 1:
The patent introduces a sample selection model as an intermediary between the noisy sample data and the target model training process. This intermediary model filters and selects high-quality samples, thereby resolving the contradiction between maintaining training efficiency and improving training quality by eliminating noise without requiring manual review of all samples
Solution Approach 2:
The system implements self-service by automatically identifying and filtering noisy samples through the sample selection model without requiring manual annotation or review. The model autonomously processes the sample data, selects appropriate samples, and feeds them to the target model for training, thereby maintaining productivity while improving reliability
2Reliability
If manual annotation is used to obtain samples, then sample quality can be controlled, but the process is time-consuming and labor-intensive
Solution Approach 1:
The system replaces manual annotation with automated self-service through the sample selection model, which independently evaluates and selects samples based on learned patterns from historical data, thereby eliminating time-consuming manual review while maintaining sample quality
Solution Approach 2:
The sample selection model learns from historically verified high-quality samples and creates a selection criteria copy, using this learned pattern to automatically identify similar high-quality samples in the future without requiring repeated manual verification, thus reducing time loss while maintaining quality
Data Source
AI summary
A method for training a model are provided and an information recommendation system. The method includes the following. A reference training sample is obtained. A target training sample is obtained by inputting the reference training sample into a sample model. An updated target model is obtained by training a target model according to the target training sample. A target recommendation task is processed according to the updated target model, where the target recommendation task is a recommendation task for one or more target items to be recommended.