ML Taxonomy Category Suggestions for Faster Product Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual product categorization in e-commerce platforms is time-consuming and error-prone, leading to inconsistencies that are difficult to manage at scale without automated assistance.
Innovation Solution
A server or cloud service provides category suggestions using a trained machine learning model that analyzes product information to determine the highest probability node in a hierarchical taxonomy tree, refining suggestions as more details are entered, and utilizes techniques like backpropagation and class weights to mitigate imbalances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual product categorization is used by merchants, then flexibility and control over category selection are maintained, but time consumption and error rates increase significantly
Solution Approach 1:
The system enables automated self-service categorization where the machine learning model automatically assigns product categories based on product information, eliminating the need for manual merchant intervention while maintaining high accuracy through confidence thresholds and merchant override capabilities
Solution Approach 2:
The patent replaces the manual mechanical process of merchant categorization with an automated machine learning system that uses trained models to predict product categories, significantly reducing time consumption while maintaining or improving categorization accuracy
2Productivity
If automated category suggestions are implemented, then efficiency and consistency are improved, but system complexity and potential errors from algorithm limitations are introduced
Solution Approach 1:
The system segments the categorization process into distinct components: machine learning model for prediction, confidence threshold filtering, and merchant review/override capability. This modular approach manages complexity by separating automated and manual decision-making functions
Solution Approach 2:
The patent introduces an intermediary layer between automated ML predictions and final category assignment, where merchant review and override capabilities act as a buffer to catch algorithm errors while still benefiting from automated efficiency in most cases
3Measurement precision
If deep hierarchical taxonomy is used for product categorization, then categorization precision and product grouping accuracy are improved, but difficulty in selection and processing increases
Solution Approach 1:
The machine learning model performs preliminary analysis of product information and pre-determines the most likely category path through the hierarchical taxonomy, presenting merchants with a small set of high-probability options rather than requiring navigation through the entire deep hierarchy
Solution Approach 2:
The system dynamically adjusts the confidence threshold parameter based on the depth of the taxonomy hierarchy and the quality of product information available, allowing the model to select between broader and more specific categories appropriately while managing selection difficulty
4Stability of the object's composition
If consistent categorization rules are enforced across all products, then data quality and analytics reliability are improved, but flexibility in handling diverse product types and edge cases is reduced
Solution Approach 1:
The system dynamically adapts to different product types and edge cases by allowing merchant override of automated suggestions, combining consistent automated rules with flexible case-by-case adjustments to maintain both consistency and adaptability
Solution Approach 2:
The patent implements feedback mechanisms where merchant corrections and overrides are used to retrain and improve the machine learning model, enabling the system to learn from diverse edge cases and improve its ability to handle varied product types while maintaining overall consistency
Data Source
AI summary
A method for categorizing a product, the method including receiving information for the product; inputting the information into a trained machine learning model for a taxonomy tree; receiving a plurality of arrays, each array representing a level in the taxonomy tree and consisting of probabilities for each category represented in the level that the product is categorized in that category; choosing, from a highest level tier array, a category having a highest probability, thereby designating a tier prediction; collecting, from a second level tier array, all children of the tier prediction; determining whether a highest probability from the children of the tier prediction exceeds a threshold, and if yes, choosing the category with the highest probably as a new tier prediction; and repeating the determining; when the threshold is not exceeded or if the tier prediction has no children, and selecting the tier prediction as a predicted category.