Machine Learning Post-Filtering for Named Entity Recognition Accuracy

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing named entity recognition systems face challenges in accurately identifying named entities due to static heuristics that fail to incorporate contextual information, leading to false positives and requiring extensive redevelopment for different document types.

Innovation Solution

A machine learning model is used to identify candidate named entities by extracting chunks of text, creating feature vectors that include contextual relationships, and analyzing these vectors to correct errors, allowing for automatic updating and integration of new features relevant to specific document types.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If static heuristics are used for named entity recognition, then the system is simple and easy to implement, but the accuracy deteriorates due to inability to incorporate contextual information

Engineering Contradiction:

Improvenamed entity recognition accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces static heuristic rules with a machine learning-based post-filtering system that uses neural networks to analyze contextual information around candidate named entities, enabling dynamic decision-making based on learned patterns from training data

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system transforms fixed heuristic parameters into adaptive machine learning models that can adjust their decision boundaries based on contextual features, allowing the system to learn optimal filtering criteria from labeled training data

Inventive Principle:

Principle #35Parameter changes

2Adaptability or versatility

If static heuristics are used for named entity recognition, then the system requires minimal updates, but adaptability to different document types deteriorates

Engineering Contradiction:

Improveadaptability to different document typesVSAvoidtime for manual updates

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system performs self-training by automatically learning from labeled training data and updating its own parameters, eliminating the need for manual program updates when adapting to new document types

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent implements a dynamic post-filtering system that can be retrained on new data to adapt to different document types, transitioning from static rules to a flexible, learnable system that evolves with new requirements

Inventive Principle:

Principle #15Dynamics

3Adaptability or versatility

If extensive redevelopment is performed for different document types, then adaptability improves, but productivity deteriorates due to time-consuming updates

Engineering Contradiction:

Improveadaptability to different document typesVSAvoidprocessing speed

Core Design Contradiction:

Adaptability or versatilityVSProductivity

Solution Approach 1:

The system pre-trains on a comprehensive training set to learn general patterns, then uses this pre-trained knowledge to quickly adapt to specific document types through targeted retraining on smaller, type-specific datasets, avoiding the need for complete system redevelopment

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11687719B2Post-filtering of named entities with machine learning

Publication Date: 2023.06.27 LEVERTON HOLDING LLC

AI summary

A method for identifying errors associated with named entity recognition includes recognizing a candidate named entity within a text and extracting a chunk from the text containing the candidate named entity. The method further includes creating a feature vector associated with the chunk and analyzing the feature vector for an indication of an error associated with the candidate named entity. The method also includes correcting the error associated with the candidate named entity.