Adaptive Node Collapsing for Hierarchical Classification Ontologies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Hierarchical classification models face challenges in handling long-tail distributions and semantic diversity in datasets, leading to non-comprehensible errors and difficulties in learning fine-grained semantic differences, especially in medical coding tasks where data distribution is skewed and insufficient training data exists for rare events.
Innovation Solution
A mechanism is introduced to find an optimal surface in the hierarchical classification ontology, balancing accuracy and coverage by determining an abstractness level that allows the model to classify instances to parent nodes when leaf nodes are difficult to train, using a neural network architecture with an indicative layer to identify nodes that can be collapsed adaptively, thereby reducing the complexity of the classification task.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the model trains on long-tail distributed data with fine-grained leaf nodes, then classification coverage is improved, but training accuracy deteriorates due to insufficient data for rare events
Solution Approach 1:
The patent implements dynamic node collapsing where the classification hierarchy is adaptively modified during training based on data distribution characteristics. Nodes are collapsed when insufficient training data is detected, allowing the model to dynamically adjust the depth of the hierarchy and focus computational resources on well-represented classes while maintaining coverage of rare events at higher levels.
Solution Approach 2:
The patent changes the structural parameters of the classification hierarchy by collapsing nodes at different depths based on data availability. This transforms the fixed ontology into a variable structure where the effective classification depth varies across different regions of the hierarchy, allowing optimal training accuracy for each local region while maintaining overall system adaptability.
2Measurement precision
If the model attempts to classify all leaf nodes, then classification specificity is improved, but computational complexity increases due to the need to learn fine-grained semantic differences
Solution Approach 1:
The patent extracts and removes difficult-to-learn fine-grained leaf nodes from the active classification path when insufficient training data is detected. This creates a pruned hierarchy where only the most learnable and data-supported nodes are actively classified, reducing computational complexity while maintaining specificity for well-represented classes.
Solution Approach 2:
The patent segments the classification task into manageable regions of the hierarchy based on data distribution. Instead of uniformly learning all leaf nodes, the system identifies and segments the hierarchy into regions with sufficient data for fine-grained classification and regions where higher-level abstraction is more appropriate, thereby reducing overall computational complexity.
3Device complexity
If the model uses fixed hierarchical structure, then system simplicity is maintained, but adaptability to long-tail distributions deteriorates
Solution Approach 1:
The patent transforms the fixed hierarchical structure into a dynamic one that automatically adapts to long-tail distributions. The system monitors training data characteristics and dynamically collapses nodes in regions with insufficient data, allowing the hierarchy to self-adjust to the actual data distribution without requiring manual intervention or complex pre-processing.
Solution Approach 2:
The system performs self-adjustment by automatically detecting data distribution patterns and initiating node collapsing operations based on training performance metrics. This self-service mechanism eliminates the need for external intervention to handle long-tail distributions, maintaining system simplicity while achieving adaptability through automated hierarchical reconfiguration.
Data Source
AI summary
A mechanism is provided in a data processing system having a processor and a memory. The memory comprises instructions which are executed by the processor to cause the processor to implement a training system for finding an optimal surface for hierarchical classification task on an ontology. The training system receives a training data set and a hierarchical classification ontology data structure. The training system generates a neural network architecture based on the training data set and the hierarchical classification ontology data structure. The neural network architecture comprises an indicative layer, a parent tier (PT) output and a lower leaf tier (LLT) output. The training system trains the neural network architecture to classify the training data set to leaf nodes at the LLT output and parent nodes at the PT output. The indicative layer in the neural network architecture determines a surface that passes through each path from a root to a leaf node in the hierarchical ontology data structure. The training system trains a classifier model for a cognitive system using the surface and the training data set.