Generative AI Model for E-commerce Size Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
E-commerce systems face challenges in providing accurate size recommendations due to non-standardization of item sizes, data sparsity, and the 'cold start' problem, leading to poor user experiences and increased return rates, which result in inefficiencies and environmental impact.
Innovation Solution
A generative machine learning model using an ordinal regression-based Bayesian approach to estimate true sizes of users and items, leveraging past transactions, measurement data, and user-provided information, while handling data sparsity and multiple personas, to provide personalized fit recommendations with confidence intervals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional recommendation systems are used without standardized size data, then system simplicity is maintained, but recommendation accuracy deteriorates leading to poor user experiences and increased return rates
Solution Approach 1:
The patent introduces size standardization data as an intermediary layer between user measurements and item dimensions. This mediator translates non-standardized item sizes into standardized measurements, enabling accurate recommendations without requiring complete standardization of all product data. The intermediary handles the complexity of size variation across different brands and items.
Solution Approach 2:
The system transforms size recommendations from categorical labels (S, M, L, XL) to continuous measurement parameters (inches, centimeters). This parameter change enables precise matching between user body measurements and item dimensions, improving recommendation accuracy while allowing the system to work with both standardized and non-standardized data through mathematical transformations.
2Reliability
If multiple size options are shipped to ensure fit, then user satisfaction is improved, but packaging waste and transportation emissions increase
Solution Approach 1:
The system performs preliminary size determination by calculating recommended sizes and confidence scores before shipping. By pre-determining the most likely fitting size using Bayesian analysis and size standardization, the system enables users to order a single size with high confidence, eliminating the need to ship multiple size options and thereby reducing packaging waste and transportation emissions.
Solution Approach 2:
The system incorporates feedback loops where return data and user fit feedback are continuously used to refine size recommendations. This feedback mechanism improves recommendation reliability over time, allowing the system to become more confident in single-size recommendations and further reduce the need for multiple shipments and associated waste.
3Measurement precision
If Bayesian analysis with posterior distribution calculation is performed, then parameter inference accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the Bayesian analysis into modular components: likelihood calculation based on size standardization, prior distribution definition, and posterior computation. By breaking down the complex Bayesian inference into separate calculable modules, the system achieves accurate parameter estimation while managing computational complexity through structured, incremental processing.
Solution Approach 2:
The system transforms the Bayesian parameter estimation problem from inferring discrete size categories to estimating continuous measurement parameters. This parameter transformation simplifies the mathematical operations required for posterior calculation, as continuous parameters allow for more efficient approximation methods compared to discrete categorical inference.
Data Source
AI summary
Some aspects of the present disclosure relate to computer processes for generating and training a generative machine learning model to estimate the true sizes of items and users of an electronic catalog and subsequently applied to determine fit recommendations, as well as confidence values for the fit recommendations, for how a particular item may fit a particular user. During training, the disclosed generative model can implement Bayesian statistical inference to calculate estimated true sizes of both items and users of an electronic catalog using both (1) a prior distribution of sizes for items and users and (2) a distribution based on obtained evidence regarding how items actually fit users. The resulting posterior distribution can be approximated using a proposal distribution used to generate the fit recommendations and associated confidence values.