Named Entity Span Detection for Complex Range Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for named entity recognition fail to provide sufficiently high accuracy in recognizing the ranges of named entities, leading to reduced accuracy in main natural language processing tasks, and are often specialized for specific domains with limited training data, complicating their structure and performance.
Innovation Solution
A machine learning model is developed to separately recognize the range of named entities using training data that includes text, class information, and position information, allowing for improved accuracy and versatility across different domains by simplifying the recognition process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing machine learning models for named entity recognition are used, then the recognition process can be performed, but the accuracy in recognizing the ranges of named entities is insufficient
Solution Approach 1:
The patent segments the named entity recognition task into two independent sub-tasks: (1) identifying candidate spans using a span detector that considers class information and position information, and (2) classifying these spans using a classifier. This segmentation allows each component to be optimized independently, improving overall recognition accuracy while maintaining system reliability through modular design.
Solution Approach 2:
The patent creates a universal framework that can handle various types of named entities (discontinuous, overlapping, nested) through a single integrated system. The span detector and classifier work together to universally process different entity types without requiring domain-specific model modifications, thereby improving both accuracy across domains and reliability of the recognition process.
2Measurement precision
If machine learning models are specialized for specific domains with limited training data, then domain-specific accuracy may improve, but the model structure becomes more complex
Solution Approach 1:
The patent develops a universal span detector and classifier framework that can be applied across multiple domains without requiring complex domain-specific adaptations. The model accepts class information and position information as inputs and produces accurate span predictions regardless of the domain, thereby maintaining simplicity while achieving domain-specific accuracy through flexible input handling rather than structural complexity.
Solution Approach 2:
The patent changes the parameters fed into the model (class information and position information) rather than changing the model structure itself to adapt to different domains. This allows the same simple model structure to achieve domain-specific accuracy by receiving domain-relevant parameter inputs, avoiding the need for complex domain-specific model variants.
3Adaptability or versatility
If traditional named entity recognition approaches are used, then the process can be completed, but the handling of complex entities (discontinuous, overlapping, nested) becomes difficult
Solution Approach 1:
The patent segments the complex entity handling problem into span detection and classification stages. The span detector generates candidate spans based on class information and position information, and the classifier accurately determines the final spans. This segmentation enables the system to accurately handle discontinuous, overlapping, and nested entities by processing them through distinct computational stages rather than attempting to solve all complexities in a single step.
Solution Approach 2:
The patent introduces an intermediary classification stage between span candidate generation and final output. The classifier acts as a mediator that refines the candidate spans produced by the span detector, accurately resolving complex entity configurations (discontinuous, overlapping, nested) through learned patterns. This intermediary step significantly improves recognition accuracy for complex entities while maintaining system versatility.
Data Source
AI summary
A computer acquires training data including first text, first class information indicating a class mapped to a single word contained in the first text, first position information indicating a position of the single word in the first text, and first range information indicating a range of a first named entity that includes the single word in the first text. The computer executes, based on the training data, machine learning of a machine learning model which is used to estimate, from text, class information, and position information, range information of a named entity included in the text.