Model training and recognition method and device for named entity recognition and storage medium

By combining rule bases, deep learning, and clustering methods, this method annotates and predicts media resource data, solving the problem of entity name recognition in the media resource field and achieving efficient entity name recognition and model optimization.

CN116911299BActive Publication Date: 2026-07-10CHINA MOBILE COMM LTD RES INST +1
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
CHINA MOBILE COMM LTD RES INST
Filing Date
2023-02-28
Publication Date
2026-07-10

AI Technical Summary

Technical Problem

Existing named entity recognition technologies in the media resource field cannot effectively identify entity names due to rapid data updates, lack of labeled data, and missing features. Rule-based methods cannot cover all grammatical rules, deep learning-based methods cannot learn, and clustering-based methods perform poorly.

Method used

By combining rule base, deep learning, and clustering methods, the data samples in the training set are labeled and predicted. The prediction results are calibrated through clustering until the convergence condition is met, and the NER model is optimized.

Benefits of technology

It improves the performance of entity name recognition in the media resource domain, reduces the cost of manual annotation, and improves recognition accuracy and model training effect.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116911299B_ABST
    Figure CN116911299B_ABST
Patent Text Reader

Abstract

The application discloses a model training and recognition method and device for named entity recognition and a storage medium. The training method comprises the following steps: labeling data samples in a training set based on a rule base to obtain first labeled names of the data samples; training an initial NER model based on the data samples and corresponding first labeled names; predicting the data samples in the training set based on the trained NER model, clustering each data sample based on a predicted entity name and an intermediate result, and determining second labeled names of each data sample based on a clustering result; and continuing to train the NER model based on the data samples and corresponding second labeled names until the clustering result meets a convergence condition, so that a trained NER model is obtained. The automatic labeling of samples can be realized by using the rule base, the prediction result can be calibrated and the subsequent model can be optimized and trained based on clustering, and the training effect of the NER model can be effectively improved.
Need to check novelty before this filing date? Find Prior Art