Ontology Population with User Refinement and Deep Learning

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional ontology population systems lack accuracy and fail to consider the hierarchical structure of ontologies, resulting in insufficient performance in real-world business domains, and do not allow for user-centric refinement to adapt to changing conceptualizations.

Innovation Solution

A method that involves determining candidate ontologies for alignment from multiple knowledge bases, using deep learning hierarchical classification to align concepts between the initial and target ontologies, and allowing users to build, change, and grow their ontologies with human-in-the-loop refinement, incorporating new facts from unstructured data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional ontology population systems are used, then the process is automated, but the accuracy and performance are insufficient

Engineering Contradiction:

Improveontology population accuracyVSAvoidmanual refinement requirement

Core Design Contradiction:

Measurement precisionVSExtent of automation

Solution Approach 1:

The system implements feedback loops where user corrections and refinements are incorporated back into the ontology population process. The hierarchical classification model is trained on user-provided corrections, allowing the system to learn from feedback and improve accuracy over time while maintaining automated operation.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The ontology population system transitions from a static automated process to a dynamic adaptive system. The classification model dynamically adjusts its parameters and decision boundaries based on user refinements and feedback, enabling it to adapt to domain-specific nuances and improve accuracy progressively.

Inventive Principle:

Principle #15Dynamics

2Reliability

If existing ontology alignment methods are used, then existing knowledge can be reused, but the hierarchical structure is not considered

Engineering Contradiction:

Improveknowledge reuse capabilityVSAvoidhierarchical structure handling

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The ontology alignment process is segmented into hierarchical levels, where the system processes and aligns concepts at different depths of the ontology hierarchy separately. This segmentation allows the system to handle the complex hierarchical structure systematically while maintaining the ability to reuse existing knowledge from source ontologies.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system adds a hierarchical dimension to the ontology alignment process by organizing concepts into multi-level hierarchies. This dimensional approach allows the system to capture structural relationships and dependencies that flat alignment methods miss, improving reliability in knowledge reuse while managing complexity through structured organization.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

3Adaptability or versatility

If user refinement is incorporated, then adaptability to changing conceptualizations improves, but system complexity increases

Engineering Contradiction:

Improveconceptualization adaptabilityVSAvoidsystem architecture complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system employs a nested architecture where user refinement interfaces are embedded within the automated classification pipeline. The hierarchical classification model is nested within the ontology population system, which itself is nested within the broader knowledge management ecosystem. This nesting allows user refinement to be integrated seamlessly without proportionally increasing overall system complexity.

Inventive Principle:

Principle #7Nested doll (Nesting)

Solution Approach 2:

The system introduces intermediary components that mediate between user refinements and the core classification algorithm. These intermediaries process, validate, and translate user feedback into format suitable for model retraining, reducing the complexity burden on the core system while maintaining high adaptability to changing conceptualizations.

Inventive Principle:

Principle #24Intermediary (Mediator)

4Measurement precision

If deep learning hierarchical classification is used, then alignment accuracy improves, but computational requirements increase

Engineering Contradiction:

Improveconcept alignment accuracyVSAvoidcomputational resource consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system applies partial deep learning hierarchical classification by focusing computational resources on the most critical alignment decisions and hierarchical levels. Rather than applying full deep learning uniformly across all concepts, the system selectively applies sophisticated classification where it provides maximum benefit, reducing overall computational requirements while maintaining high accuracy for key alignments.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS11593419B2User-centric ontology population with user refinement

Publication Date: 2023.02.28 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

One embodiment provides a method that includes determining candidate ontologies for alignment from multiple available knowledge bases. An initial target ontology is selected from the candidate ontologies and correcting the initial selected ontology with received refinement input. Concepts in the selected initial ontology are aligned with concepts of the target ontology using a deep learning hierarchical classification with received review input. A user is assisted to build, change and grow the selected initial ontology exploiting both the target ontology and new facts extracted from unstructured data.