Entity Recognition Accuracy via Dictionary Verification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current named entity recognition technologies suffer from inaccuracies, including omissions and errors, with an average accuracy rate of 85%, which affects the reliability of natural language processing applications such as knowledge domain mapping and information extraction.

Innovation Solution

An entity recognition method that involves acquiring and updating an entity dictionary based on historical recognition results, verifying recognition results against the dictionary, and correcting or updating the dictionary to improve accuracy, particularly for entities like drug ingredients with rare words or special characters, using modules for acquisition, verification, and correction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional named entity recognition algorithms are used, then processing speed is maintained at acceptable levels, but recognition accuracy reaches only 85% with omissions and errors

Engineering Contradiction:

Improveentity recognition accuracyVSAvoidrecognition system complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent implements a feedback mechanism where recognition results are verified against an entity dictionary, and errors are fed back to correct the dictionary and retrain the model. This iterative feedback loop continuously improves recognition accuracy from the initial 85% toward higher precision by learning from mistakes.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent pre-compiles an entity dictionary from historical recognition results before processing new text. This preliminary action of organizing historical data into a structured dictionary enables faster and more accurate verification during recognition, improving precision without significantly increasing real-time processing complexity.

Inventive Principle:

Principle #10Preliminary action

2Adaptability or versatility

If entity recognition is performed on complex entities with rare words or special characters, then coverage is improved, but error rates increase

Engineering Contradiction:

Improveentity coverageVSAvoidrecognition accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system performs self-service by automatically verifying recognition results against the entity dictionary and self-correcting errors. When complex entities with rare words or special characters are encountered, the system automatically detects mismatches, updates the dictionary, and retrains the model without human intervention, maintaining high accuracy across diverse entity types.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces purely algorithmic recognition with a hybrid system that incorporates dictionary-based verification and rule-based checks (such as character sequence validation). This substitution of pure AI with structured verification mechanisms improves accuracy for complex entities by cross-referencing against curated knowledge.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Measurement precision

If historical recognition results are used to compile entity dictionary, then recognition accuracy is improved, but data processing time increases

Engineering Contradiction:

Improverecognition accuracyVSAvoiddata processing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The entity dictionary is compiled from historical recognition results as a preliminary action before processing new text. By pre-organizing historical data into a structured dictionary format, the system enables fast lookup and verification during actual recognition tasks, reducing real-time processing time while maintaining high accuracy.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent creates a copy of historical recognition results in the form of an entity dictionary, separating the historical data from the active processing stream. This copying allows the system to leverage historical accuracy information without reprocessing all historical data during each new recognition task, thus improving accuracy without proportionally increasing processing time.

Inventive Principle:

Principle #26Copying

Data Source

PatentUS20230177271A1Entity recognition methods and apparatuses, electronic devices and storage media

Publication Date: 2023.06.08 BOE TECHNOLOGY GROUP CO LTD

AI summary

The present application discloses an entity recognition method, an entity recognition apparatus (110), an electronic device (100) and a readable storage medium. The method includes: acquiring an entity dictionary (S10), where the entity dictionary is compiled according to historical recognition results from an entity recognition model; verifying an entity recognition result from the entity recognition model according to the entity dictionary and a text to be recognized (S20); and in response to determining that a verification result is abnormal, updating the entity dictionary and/or correcting the entity recognition result (S30).