A method for continuous named entity recognition based on generative paradigm

By employing a generative paradigm-based continuous named entity recognition method, utilizing instruction learning and the T5 language model of the PyTorch framework, combined with EWC regularization and anomaly behavior judgment, the method solves the problems of recognition accuracy and adaptability of traditional methods under large-scale complex data, and achieves efficient and accurate entity recognition and anomaly response.

CN120524950BActive Publication Date: 2026-06-23JIANGSU TONGXINGBAO INTELLIGENT TRANSPORTATION TECH CO LTD
2 Cites 0 Cited by

Patent Information

Application Number
CN202510606988.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-05-12
Publication Date
2026-06-23
Estimated Expiration
2045-05-12

AI Technical Summary

Technical Problem

Traditional continuous named entity recognition methods are inefficient and inaccurate when dealing with large-scale, highly complex data. In particular, they are difficult to accurately label when faced with newly emerging entity types or rare entity instances, resulting in a decrease in recognition accuracy.

Method used

A generative paradigm-based approach is adopted, which constructs a task framework through instruction learning technology, trains the model using the PyTorch framework and T5 language model, and combines EWC regularization terms and representativeness difficulty sample replay methods to construct a normal behavior benchmark and an abnormal behavior judgment system, sets clear abnormal behavior judgment rules, and monitors and triggers an early warning mechanism in real time.

Benefits of technology

It improves the model's ability to process complex and diverse texts, alleviates the problem of catastrophic forgetting, enhances the accuracy and stability of entity recognition, and can respond to abnormal situations in a timely manner and take effective measures to adapt to the entity recognition needs of various text scenarios.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The application discloses a kind of based on the continuous named entity recognition method of generative paradigm, it is related to artificial intelligence algorithm technical field.The application includes constructing generative task framework, using Instruction learning technology, according to different NER application scene and demand custom task instruction, according to general entity category and specific field requirement flexible adjustment entity option, obtain input sentence from the text data source of webpage text, document library, social media text, determine output entity format as [(entity type 1: entity mention 1), (entity type 2: entity mention 2)... ], to guide generative model to complete NER task.By using Instruction learning technology to construct generative task framework, can customize task instruction according to different NER application scene, adjust entity option, guide T5 language model to convert traditional task into generative task, enhance the flexibility and practicality of model processing different tasks, so that it can better adapt to various text scenes, improve the processing capacity to complex and various text.
Need to check novelty before this filing date? Find Prior Art