Hierarchical Product Classification Using Unified Multi-Layer Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning methods for hierarchical product classification are inefficient due to the need to train separate models for each level of the hierarchy, leading to increased training time and computational intensity, and often require retraining thousands of models, which is cumbersome and costly.
Innovation Solution
A multi-layered network model that defines a domain embedding using a word embedding layer and expresses the hierarchy in a bitmap structure, allowing a single model to predict multiple levels of the hierarchy by using a parent function and child function based on dot products, eliminating the need for level-wise retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate models are trained for each level of the hierarchy, then the classification can be performed level-by-level, but the training time and computational complexity increase significantly
Solution Approach 1:
The patent combines multiple separate classification models into a single unified model that processes all hierarchy levels simultaneously. The system uses a shared embedding layer that generates representations for all levels at once, eliminating the need to train and maintain N separate models. This merging approach reduces training time while maintaining classification accuracy across all hierarchy levels.
Solution Approach 2:
The patent creates a universal model that performs classification across multiple hierarchy levels using a single architecture. The model employs a shared backbone network that can predict classifications at different levels of the hierarchy simultaneously, making the system multi-functional rather than requiring separate specialized models for each level.
2Measurement precision
If separate models are trained for each level of the hierarchy, then each level can be optimized independently, but the maintenance and orchestration of N models becomes complex
Solution Approach 1:
The patent merges N separate models into a single unified model architecture. The system uses shared layers and parameters across all hierarchy levels, reducing the number of models from N to 1. This simplifies maintenance, deployment, and orchestration while preserving the ability to perform accurate classification at each level through the shared representation learning.
Solution Approach 2:
The patent segments the classification task into hierarchical levels but processes them within a unified model framework. The model uses shared embedding layers that generate level-specific representations internally, allowing each level to be optimized through the same architecture rather than requiring separate model maintenance for each segment.
3Adaptability or versatility
If the model learns the hierarchy from data, then the hierarchy can be adaptive, but the training time increases significantly to achieve accuracy
Solution Approach 1:
The patent incorporates predefined hierarchy structures into the model architecture before training begins. The hierarchy information is embedded in the model's structure and parameters, allowing the model to leverage this prior knowledge during training. This preliminary incorporation of hierarchy knowledge reduces the time needed to learn hierarchical relationships from scratch while maintaining adaptability to the specific domain hierarchy.
Data Source
AI summary
This disclosure relates generally to system and method for hierarchical category classification of products. Generally in supervised hierarchical classification, the hierarchy structure is predefined. However, majority of the current machine learning methods either expect the model to learn the hierarchy from the data or requires separate models trained at each level taking the prediction of previous level as an additional input, thereby increasing latency in achieving training accuracy and/or requiring an explicit maintenance module to orchestrate inference and retrain multiple models (corresponding to the number of levels in the hierarchy). The disclosed method and system allows the predefined knowledge about hierarchy drive the learning process of a single model, which predicts all levels of the hierarchy. The disclosed multi-layer network model arrives at a consensus based on prediction at each level, thereby increasing the accuracy of prediction and reducing the training time.