Hierarchical Industry Classification Using Neural Network Sub-Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The manual classification of business entities into industry classifications is labor-intensive and prone to errors, and automated systems face inaccuracies due to the large number of classifications and imbalances in data, leading to incomplete and unreliable business profile data.
Innovation Solution
A computer-implemented classification system using a neural network with a word-based sub-classifier and a category-based sub-classifier, along with a meta-classifier, to generate vector representations of business summaries for accurate classification against a hierarchical industry classification structure, addressing underrepresented parent industry classifications and optimizing weight values for improved predictive capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated classification systems are used to classify business entities into industry classifications, then productivity is improved, but measurement precision deteriorates due to the large number of classifications and data imbalances
Solution Approach 1:
The classification system is segmented into multiple specialized classifiers: a parent industry classifier that handles high-level categories with sufficient data, and child industry classifiers that handle specific sub-categories. Each classifier is trained on data appropriate to its level, with the parent classifier processing summaries at a general level and child classifiers refining classifications for specific industries. This segmentation allows the system to maintain high productivity while improving precision by avoiding the extreme classification problem of directly classifying into all 2000+ NAICS codes.
Solution Approach 2:
Parent industry classifications serve as intermediary categories between the business summary and specific child industry classifications. The system first classifies into parent industries (which have sufficient training data), then uses these parent classifications as intermediaries to guide subsequent child industry classification. This intermediary approach resolves the data imbalance problem by using well-represented parent categories to bridge the gap to less-represented child categories.
2Measurement precision
If manual classification is used to ensure accuracy, then measurement precision is improved, but productivity deteriorates due to labor intensity
Solution Approach 1:
The classification system performs self-service by automatically classifying business summaries into industry categories without requiring manual human intervention. The multi-classifier system with parent and child industry classifiers autonomously processes summaries, using machine learning models trained on available data to make accurate classifications. This self-service capability maintains high accuracy while achieving the productivity benefits of automation.
Solution Approach 2:
The system incorporates feedback mechanisms where classification results are continuously evaluated and used to improve future classifications. The parent-child classifier structure provides feedback loops where parent industry classification accuracy informs child industry classification, and performance metrics from real-world usage feed back into model training and refinement, progressively improving precision while maintaining automated productivity.
3Productivity
If automated classification is implemented without addressing data imbalances, then productivity is improved, but reliability deteriorates due to incomplete and unreliable business profile data
Solution Approach 1:
The system changes the parameter of classification granularity by operating at two different levels: parent industry level and child industry level. By adjusting this granularity parameter, the system can adapt to data availability - using parent classifications when child classification data is insufficient, and refining to child classifications when data supports it. This parameter change approach maintains reliability by avoiding classifications where data is too incomplete while preserving productivity through automated processing.
Data Source
AI summary
A classification system is provided for classifying text-based business summaries, referred to herein as “summaries,” against a hierarchical industry classification structure. The classification system includes a word-based sub classifier that uses a neural network to generate a vector space for each summary in a training set, where each summary in the training set is known to correspond to a particular industry classification in the hierarchical industry classification structure. Weight values in the hidden layer of a neural network used by the word-based sub classifier are changed to improve the predictive capabilities of the neural network in the business summary classification context. Embodiments include increasing representation in the training set for underrepresented parent industry classifications and attributes of the hierarchical industry classification structure, such as distances between industry classifications and whether industry classifications are in the same subgraph. The completion of training of the word-based sub classifier is based upon whether a performance metric, such as an hF1 score, satisfies one or more early stopping criteria. The classification system also includes a category-based sub classifier and a meta classifier.