AI-Based Named Entity Recognition for Multi-Nested Entities
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing named entity recognition (NER) technologies face inefficiencies and inflexibilities in handling multi-nested and multi-category named entities, particularly due to the difficulty in designing decoding rules for a large number of entity categories.
Innovation Solution
An AI-based NER method and apparatus that performs vector transformation, integration, and classification processing on text elements to construct candidate entity phrases, using a scan cycle to manage scan quantities and determine category affiliations, enhancing efficiency and precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional NER decoding rules are used to handle multiple entity categories, then the system can recognize named entities, but the complexity of designing and managing decoding rules increases significantly
Solution Approach 1:
The patent replaces the mechanical decoding rule system with a neural network-based semantic analysis system. Instead of using complex if-else decoding rules to determine entity categories, the system employs a neural network model that automatically learns and predicts entity categories based on contextual semantic understanding, thereby reducing rule complexity while maintaining multi-category recognition capability
Solution Approach 2:
The patent changes the parameter representation from discrete decoding rules to continuous neural network parameters. The system transforms the categorical decoding problem into a parameter-based prediction problem where the neural network learns optimal parameters for category classification, enabling flexible handling of multiple entity categories without explicit rule management
2Reliability
If comprehensive candidate entity phrases are generated to ensure high recall, then the recognition coverage improves, but the processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary actions by pre-processing text elements and their representations before the main recognition process. The system prepares candidate entity phrases and their corresponding text representations in advance, organizing them in a structured format that enables efficient subsequent processing and reduces computational overhead during the actual recognition phase
Solution Approach 2:
The patent segments the NER process into distinct stages: text element extraction, candidate entity phrase generation, text representation construction, and category prediction. This segmentation allows each stage to be optimized independently, processing only relevant information at each step rather than handling all data uniformly, thereby reducing overall processing time while maintaining comprehensive recognition
Data Source
AI summary
Aspects of this disclosure are directed to an artificial intelligence (AI)-based named entity recognition (NER) method and apparatus, an electronic device, and a non-transitory computer-readable storage medium. The method can include performing, by an electronic device, vector transformation processing on text elements in a to-be-recognized text to obtain text representations of the text elements, and constructing a candidate entity phrase according to text elements that are in the to-be-recognized text and whose total quantity does not exceed an element quantity threshold. The method can further include performing integration processing on text representations corresponding to the text elements in the candidate entity phrase to obtain a text representation of the candidate entity phrase, and performing classification processing on the text representation of the candidate entity phrase to determine a category to which the candidate entity phrase belongs in a non-named entity category and a plurality of named entity categories.