Domain Named Entity Recognition via Multi-Stage Tagging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current named entity recognition methods in Chinese texts face challenges due to incorrect word segmentation and reliance on complete dictionaries or rules, leading to inaccurate recognition of proprietary named entities, especially in changing contexts.
Innovation Solution
A method and device for recognizing domain named entities that involves word segmentation followed by tagging with a domain-specific tag set, including location tags, and extraction based on domain named entity extraction rules, reducing reliance on dictionary matches and improving boundary accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If word segmentation is conducted on Chinese text, then the text can be processed for entity recognition, but incorrect word segmentation leads to inaccurate determination of named entity boundaries
Solution Approach 1:
The patent segments the named entity recognition task into multiple tagging stages (first tagging for entity identification, second tagging for boundary determination). This multi-stage segmentation allows the system to handle word segmentation errors by progressively refining entity boundary identification through separate tagging passes, rather than relying on a single segmentation result.
Solution Approach 2:
The patent introduces tagging as an intermediary step between word segmentation and final entity extraction. The tagging process acts as a mediator that can correct or adjust for word segmentation errors by independently identifying entity boundaries through linguistic patterns and context, rather than directly relying on segmented word boundaries.
2Ease of manufacture
If dictionary or rule-based methods are used for named entity recognition, then the process is simple to implement, but the accuracy completely depends on the completeness of the dictionary or rule
Solution Approach 1:
The patent employs dynamic tagging rules that can adapt to different contexts and entity types. Rather than using static dictionary matches, the system applies configurable tagging patterns that can be adjusted based on the specific domain and entity characteristics, allowing the recognition accuracy to improve with domain knowledge while maintaining systematic implementation.
Solution Approach 2:
The patent changes the parameters of entity recognition from fixed dictionary matches to flexible tagging patterns with adjustable parameters. The tagging system allows modification of entity boundaries, types, and attributes based on contextual parameters, enabling accurate recognition of new or changing entities without requiring complete dictionary updates.
3Reliability
If complete dictionary coverage is maintained for all entities, then all entities can be recognized accurately, but the system cannot adapt to changing entity coverage
Solution Approach 1:
The patent creates a universal tagging system that can handle multiple entity types and domains through a unified set of tagging patterns. The tagging framework is designed to be domain-agnostic yet adaptable, allowing the same basic tagging mechanism to recognize various entities (persons, organizations, locations, domain-specific entities) by adjusting tagging rules rather than requiring separate dictionaries for each entity type.
Solution Approach 2:
The patent performs preliminary tagging of all text elements with potential entity characteristics before final entity determination. This preliminary action tags all possible entities with provisional labels, allowing the system to later refine or adjust entity identification based on complete context analysis, thereby adapting to changing entity coverage without missing potential entities.
Data Source
AI summary
The invention discloses a method and device for recognizing a domain named entity. The main technical solution is: conducting word segmentation on a text to be recognized; tagging each segmented word in the text to be recognized according to a tag set of a domain corresponding to the text to be recognized, where the tag set includes a basic tag set based on domain named entity recognition and a tag set belongs to the corresponding domain, and the basic tag set includes a location tag of related words which constitute the domain named entity; and extracting the tagged segmented words according to a domain named entity extraction rule and constituting the domain named entity.