Variant Grouping Using Text and Image Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Online item catalogs often fail to automatically group variants of products, such as different sizes and colors, effectively, leading to inefficiencies in product representation and user experience, especially in large catalogs where human curation is impractical.
Innovation Solution
A system utilizing machine learning models, including k-nearest neighbors and elastic search approaches, to generate candidate variant items based on text and image embeddings, combined with logistic regression and Siamese networks for classification and grouping, automatically determines items to include in variant groups, suggesting these groups for human review or implementation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human curation is used to group product variants, then grouping accuracy is improved, but labor cost and time consumption increase significantly
Solution Approach 1:
The patent replaces the mechanical human curation process with an automated computer-based system that uses machine learning models (k-nearest neighbors, elastic search, logistic regression, and Siamese networks) to automatically determine variant groupings. This substitution eliminates manual labor while maintaining high grouping accuracy through sophisticated algorithms that analyze product attributes, images, and descriptions.
Solution Approach 2:
The system enables self-service automation where the computer automatically processes product data, generates candidate variant groups, and performs iterative refinement without requiring human intervention at each step. The machine learning models independently analyze and group variants based on learned patterns from training data, making the system self-sufficient for the core grouping task.
2Productivity
If automated grouping methods are used, then productivity is improved, but grouping accuracy may deteriorate
Solution Approach 1:
The system performs preliminary actions by generating multiple candidate variant groups before final determination. The k-nearest neighbors and elastic search approaches initially identify potential variants, which are then refined through logistic regression classification and Siamese network verification. This multi-stage preliminary processing ensures high accuracy while maintaining automated efficiency.
Solution Approach 2:
The patent implements feedback mechanisms where the system iteratively refines variant groupings by evaluating candidate groups against learned patterns and adjusting classifications accordingly. The machine learning models use feedback from training data and performance metrics to continuously improve grouping accuracy, ensuring automated processes achieve results comparable to or exceeding human curation.
3Measurement precision
If multiple machine learning models are combined, then grouping accuracy is improved, but system complexity increases
Solution Approach 1:
The patent segments the complex grouping task into distinct functional components handled by different machine learning models: k-nearest neighbors for initial candidate generation, elastic search for attribute-based matching, logistic regression for classification, and Siamese networks for image similarity verification. Each model handles a specific aspect of the grouping problem, making the overall system more manageable and interpretable despite the combination of multiple models.
4Productivity
If manual curation is avoided in large catalogs, then productivity is improved, but the ability to handle frequent updates deteriorates
Solution Approach 1:
The patent creates a dynamic system that can adapt to frequent catalog updates through the flexibility of machine learning models. The system can reprocess new or modified product data through the same automated pipeline, with the models adapting to new patterns in the data. This dynamic capability allows the system to handle frequent updates efficiently without requiring manual re-curation, maintaining both high productivity and adaptability.
Data Source
AI summary
A method including obtaining image data and attribute information of a first item in an item catalog. The method also can include generating candidate variant items from the item catalog for the first item using a combination of (a) a k-nearest neighbors approach to search for first candidate variant items based on text embeddings for the attribute information of the first item, and (b) an elastic search approach to search for second candidate variant items based on image embeddings for the image data of the first item. The method additionally can include performing respective classifications based on respective pairs comprising the first item and each of the candidate variant items to filter the candidate variant items. The method further can include determining a respective distance between the first item and each of the candidate variant items, as filtered. The method additionally can include determining one or more items in the candidate variant items, as filtered, to include in a variant group for the first item, based on a decision function using a predetermined threshold and the respective distance for the each of the candidate variant items, as filtered. Other embodiments are described.