Ontology-Guided Domain-Specific Training Corpus Generation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Developing domain-specific training data for natural language processing in healthcare is challenging due to the scarcity of comprehensive, domain-specific corpora, leading to models that may not generalize well and suffer from out-of-vocabulary inputs and misclassification.

Innovation Solution

A method and system for generating domain-specific training sets by extracting keywords from an ontology, querying a generic corpus to identify relevant documents, and storing them as a domain-specific training set, which can then be used to train machine learning algorithms.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If generic domain corpora are used for training, then the amount of available training data increases, but the domain-specific accuracy and vocabulary coverage deteriorates

Engineering Contradiction:

Improveamount of training dataVSAvoiddomain-specific accuracy

Core Design Contradiction:

Quantity of substanceVSMeasurement precision

Solution Approach 1:

The patent extracts domain-specific content from generic corpora by querying with ontology keywords. The system identifies and extracts relevant documents containing domain-specific terminology and concepts, separating them from the generic corpus to create a specialized training set that maintains both quantity and domain accuracy.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces an ontology database as an intermediary between generic corpora and training data. The ontology serves as a mediator that guides the selection and filtering of generic documents, enabling the system to bridge the gap between large generic datasets and domain-specific requirements through structured keyword-based querying.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If limited clinical site data is used for training, then data privacy and access control is improved, but the model generalization capability deteriorates

Engineering Contradiction:

Improvedata securityVSAvoidmodel generalization

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent creates a training corpus that serves multiple functions: it maintains the privacy benefits of limited site data while simultaneously improving model generalization. By aggregating and processing data from multiple clinical sites through ontology-based extraction, the system generates a universal training set that can be applied across different healthcare contexts without compromising data security.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent performs preliminary data aggregation and processing from multiple clinical sites before model training. By pre-processing and consolidating data from various sources into a unified ontology-based corpus, the system enables models to learn from diverse patterns upfront, improving generalization capability while maintaining the privacy advantages of not sharing raw clinical data.

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If domain-specific corpora are manually created, then the domain relevance is improved, but the time and resource consumption increases

Engineering Contradiction:

Improvedomain relevanceVSAvoidcorpus creation time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent implements a self-service approach where the system automatically queries generic corpora using ontology keywords to generate domain-specific training data. The automated querying and extraction process eliminates manual corpus creation efforts, allowing the system to serve itself by identifying and extracting relevant domain content without human intervention in the data collection phase.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent performs preliminary organization of domain knowledge into an ontology database before the actual corpus extraction. This pre-structuring of domain concepts, terms, and relationships enables efficient automated querying of generic corpora, significantly reducing the time required to create domain-specific training data compared to manual methods.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11874864B2Method and system for creating a domain-specific training corpus from generic domain corpora

Publication Date: 2024.01.16 KONINKLIJKE PHILIPS NV

AI summary

A method (100) for generating a domain-specific training set, comprising: generating (130) a generic corpus comprising a plurality of tokenized documents, comprising: (i) parsing (132) a document retrieved from the generic corpus; (ii) preprocessing (134) the parsed document; (iii) tokenizing (136) the preprocessed document; and (iv) storing (138) the tokenized document in the generic corpus; generating (140) an ontology database of tokenized entries, comprising: (i) parsing (142) an ontology entry retrieved from an ontology; (ii) preprocessing (144) the parsed entry; (iii) tokenizing (146) the preprocessed entry; and (iv) storing (148) the tokenized entry in the ontology database; querying (150), using domain-specific tokenized entries from the ontology database, the tokenized documents in the generic corpus; identifying (160), based on the query, a plurality of tokenized documents specific to the domain; and storing (170), in a training set database, the identified tokenized documents as a training set specific to the domain.