Meta-Learning Reduces Sample Selection Bias in Ecommerce Recommenders
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning-based recommender systems suffer from sample selection bias, resulting in poor prediction performance for small shops on ecommerce platforms, while performing well for large shops.
Innovation Solution
Retraining a machine-learning recommendation model using a meta-learning process to reduce sample selection bias, by identifying a sample subset of shops, creating shop-specific training datasets, calculating shop-specific and global losses, and updating the model parameters to minimize the global loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a machine learning model is trained on user-item interactions from an ecommerce platform, then the model can make predictions for shops with large sales volumes, but the model performs poorly for small shops due to sample selection bias
Solution Approach 1:
The training data is segmented into shop-specific subsets, allowing the model to learn distinct patterns for different shop categories. The system divides the overall training task into individual shop-level training tasks, each processed separately to capture shop-specific characteristics.
Solution Approach 2:
The model applies different treatment to different shops based on their characteristics. Each shop receives a customized training approach where the model learns shop-specific user-item interaction patterns, rather than applying a uniform training approach to all shops.
2Productivity
If the model is trained on all available data, then it achieves good performance for large shops, but the predictions for small shops are biased and inaccurate
Solution Approach 1:
Instead of training on all available data uniformly, the system applies partial training actions by focusing on shop-specific subsets. The model performs multiple iterations of shop-specific training, applying training actions selectively to each shop's data portion rather than processing all data in a single pass.
Solution Approach 2:
Each shop effectively trains its own model parameters through the shop-specific training process. The system allows each shop's data to independently contribute to learning shop-specific patterns, with each shop's training process serving its own prediction needs rather than relying on a globally trained model.
3Ease of manufacture
If the model uses a single global training approach, then the training process is simple and efficient, but the model cannot adapt to different shop sizes and sales volumes
Solution Approach 1:
The system implements a universal training framework that can handle multiple shop types and sizes through a single meta-learning process. The shop-specific training function serves multiple purposes: it trains individual shop models, aggregates losses across shops, and performs global parameter updates, making the training process multi-functional and adaptable to various shop characteristics.
Data Source
AI summary
The present disclosure relates to improving recommendations for small shops on an ecommerce platform while maintaining accuracy for larger shops. The improvement is achieved by retraining a machine-learning recommendation model to reduce sample selection bias using a meta-learning process. The retraining process comprises identifying a sample subset of shops on the ecommerce platform, where the sample batch includes shops in both a small size category and shops in a large size category. The machine-learning model is then used to make shop-specific user-item interaction predictions for each shop in the sample batch. A shop-specific loss is calculated for each shop based on the model's predicted user-item interactions for the shop-specific training dataset and the actual user-item interactions in the shop-specific training dataset. A global loss is calculated based on each of the shop-specific losses, and the baseline model is updated to minimize the global loss.