Ontology Matching via NLP Embedding Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital content indexing tools are typically designed to utilize a specific ontology of terms, limiting their ability to index content with keywords from different ontologies, such as proprietary or custom ontologies.
Innovation Solution
A method for ontology matching that refines a natural language processing (NLP) model to encode terms from different hierarchical ontologies as embeddings in a vector space, using triplet loss training to improve the model's understanding of hierarchical relations, and deriving a stable matching scheme to map terms from one ontology to semantically similar terms in another.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing digital content indexing tools are designed to utilize a specific ontology of terms, then the indexing can be performed efficiently with that ontology, but the ability to index content with keywords from different ontologies is limited
Solution Approach 1:
The patent introduces an intermediary NLP model that translates terms from different ontologies into a common vector space representation. This mediator enables the indexing system to handle multiple ontologies without requiring separate indexing mechanisms for each ontology, thus improving adaptability while controlling complexity through a unified approach.
Solution Approach 2:
The NLP model is designed to universally handle term encoding from various ontologies (e.g., MeSH, PubMed, custom ontologies) and map them to a common semantic space. This multi-functional capability allows a single indexing system to serve multiple ontology requirements, enhancing versatility without proportionally increasing system complexity.
2Reliability
If an NLP model is trained to understand a single primary ontology, then the model achieves high accuracy for that ontology, but it cannot effectively handle terms from different ontologies
Solution Approach 1:
The system performs preliminary training of the NLP model on a single primary ontology to establish strong foundational accuracy. Subsequently, the model is refined with additional training data from multiple ontologies, allowing it to maintain high accuracy for the primary ontology while gaining the capability to handle terms from different ontologies effectively.
Solution Approach 2:
The patent employs triplet loss training that dynamically adjusts model parameters to minimize distances between embeddings of terms from different ontologies that have similar meanings. This parameter optimization enables the model to adapt to multiple ontologies while preserving its original accuracy for the primary ontology, resolving the contradiction between specialization and generalization.
3Device complexity
If terms from different ontologies are encoded using a single NLP model, then the system remains simple, but the model cannot accurately capture hierarchical relations between terms
Solution Approach 1:
The patent implements triplet loss training that provides feedback to the NLP model based on hierarchical relations between terms. The loss function compares the distances between embeddings of terms with known hierarchical relationships and adjusts the model parameters accordingly. This feedback mechanism enables accurate capture of hierarchical relations while maintaining a relatively simple single-model architecture.
Solution Approach 2:
The system enhances the dimensional representation of terms by training the NLP model to encode not only individual term meanings but also their hierarchical relationships in the vector space. This dimensional enrichment allows the model to capture hierarchical structure information without adding complex separate processing layers, thus maintaining simplicity while improving measurement precision.
Data Source
AI summary
An ontology matching system performs operations to refine a natural language processing (NLP) model that encodes terms of a first hierarchical ontology and of a second hierarchical ontology as embeddings in a latent space. The operations include performing at least a first round of triplet loss training to decrease separation between select pairs of the embeddings sampled from the different ontologies that satisfy a first hierarchical relation while increasing separation between other pairs of the embeddings that do not satisfy the first hierarchical relation. The system then determines, from the refined NLP model, a stable matching scheme that matches each term in the first hierarchical ontology with a corresponding term of the second hierarchical ontology. Responsive to receiving terms of the first hierarchical ontology from an application, the system uses the stable matching scheme to map each of the terms to corresponding terms of the second hierarchical ontology.