Meta-Learning Reduces Sample Selection Bias in Ecommerce Recommender Systems
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 compared to large shops.
Innovation Solution
Retraining a machine-learning recommendation model using a meta-learning process to reduce sample selection bias by creating shop-specific versions of the model and adjusting global weight parameters for small and large shops.
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 items, but the model performance is significantly better for large shops than for small shops due to sample selection bias
Solution Approach 1:
The patent segments the training data by shop size (small shops vs. large shops) and creates separate processing paths for each segment. Shop identifiers are embedded to distinguish between different shop sizes, allowing the model to learn separate prediction patterns for small and large shops, thereby resolving the performance disparity between the two segments.
Solution Approach 2:
The patent applies local quality by introducing shop-size-specific embedding vectors and separate loss function calculations for small and large shops. The model adjusts its prediction behavior locally based on the shop size category, with different weightings and optimization targets for small shop items versus large shop items, improving performance for the previously underperforming small shop segment.
2Quantity of substance
If more training data is available for large shops, then the model makes better predictions for large shops, but this creates sample selection bias that degrades performance for small shops
Solution Approach 1:
The patent applies anti-weight by introducing a counterbalancing mechanism in the loss function that compensates for the imbalance in training data volume. A shop-size-based weighting factor is applied during training, giving appropriate weight to small shop predictions even though they have fewer training examples, thereby counteracting the dominance of large shop data and improving small shop prediction accuracy.
3Productivity
If the model is optimized for overall accuracy, then the global performance improves, but the performance gap between large and small shops persists due to inherent bias in the training data distribution
Solution Approach 1:
The patent implements feedback by calculating separate loss values for small shop predictions and large shop predictions, then combining them with appropriate weightings. The model receives feedback during training about its performance on each shop size category independently, allowing it to adjust its parameters to improve performance across both categories simultaneously rather than optimizing only for the dominant large shop category.
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, and then creating shop-specific versions of the recommendation model for each of the shops in the subset. Each shop-specific model is created by optimizing the baseline model to predict user-item interactions in a first training dataset for the applicable shop. Each of the shop-specific models is then tested using a second training dataset for the shop. A loss is calculated for each shop-specific model based on the model's predicted user-item interactions and the actual user-item interactions in the second training dataset for the shop. A global loss is calculated based on each of the shop-specific losses, and the baseline model is updated to minimize the global loss. The model includes small and large-shop weight parameters that are applied to user-item interaction scores and that are learned during the re-training process.