Entity Recognition Method for Nested Text Sequences
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional entity recognition methods struggle with nested entities, as they typically only recognize outermost or innermost entities, and require high computational complexity to identify all candidate entities in a text sequence.
Innovation Solution
The proposed method increases the probability of entity candidate regions covering entities in a text sequence without increasing the number of entity candidate regions, by determining entity boundary words and using them to acquire entity candidate regions through entity proposal regions and similarity calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional sequence-based annotation methods are used to recognize entities, then each word can be assigned one label, but the method cannot effectively recognize nested entities and requires high computational complexity to identify all candidate entities
Solution Approach 1:
The patent segments the entity recognition process into distinct stages: boundary word detection, entity candidate region generation, and entity recognition. This segmentation allows each stage to focus on specific tasks, improving overall accuracy while managing computational complexity through modular processing
Solution Approach 2:
The patent performs preliminary detection of boundary words before generating entity candidate regions. This preliminary action narrows down the search space early in the process, reducing the computational complexity required for subsequent entity recognition while improving accuracy by focusing on relevant regions
2Reliability
If the number of entity candidate regions is increased to improve entity coverage, then more entities can be identified, but the computational complexity increases
Solution Approach 1:
The patent applies local quality by generating entity candidate regions with different characteristics at different locations in the text. Boundary words and their contextual representations are used to create localized candidate regions that are tailored to specific areas, improving entity coverage without requiring a uniform increase in the number of candidate regions across the entire text
Solution Approach 2:
By performing preliminary boundary word detection and using context representation vectors to identify likely entity boundaries, the patent narrows down the search space before generating candidate regions. This preliminary action ensures that candidate regions are generated only where entities are likely to exist, improving coverage while controlling the total number of candidate regions and associated computational complexity
Data Source
AI summary
An entity recognition method, apparatus, electronic device, and computer-readable storage medium are provided. The method includes: determining at least one entity boundary word corresponding to a text sequence; determining at least one entity candidate region in the text sequence based on the at least one entity boundary word; and performing entity recognition on the text sequence and identifying at least one entity in the text sequence based on the at least one entity candidate region.