A text classification method and apparatus

By combining adversarial training with the similarity vector between the text representation vector and the label matrix, the problem of applying text classification models across different datasets is solved, achieving efficient model transfer and improved robustness.

CN114048290BActive Publication Date: 2025-11-14ZHONGKE DINGFU BEIJING TECH DEV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202111386639.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-11-22
Publication Date
2025-11-14
Estimated Expiration
2041-11-22

AI Technical Summary

Technical Problem

Text classification models based on deep neural networks cannot be directly applied to different datasets due to the fixed parameters and class labels, resulting in poor generalization and robustness.

Method used

By introducing adversarial training and combining the similarity vector between the text representation vector and the label matrix, the model is trained using the first and second loss functions, enabling the model to learn the common features of each category and improving the model's generalization and robustness.

Benefits of technology

It enables seamless transfer of text classification models between different datasets, avoiding the need for retraining and improving the model's generalization and robustness.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114048290B_ABST
    Figure CN114048290B_ABST
Patent Text Reader

Abstract

This application provides a text classification method and apparatus. The method includes: obtaining a similarity vector between a text representation vector and a label matrix; obtaining a first loss function from the similarity vector to the corresponding label of the text; performing a linear mapping on the text representation vector to obtain an adversarial vector of the text; obtaining a second loss function from the adversarial vector to the corresponding label of the text; combining the first and second loss functions as the total loss function of the text classification model to train the text classification model; and classifying unknown text based on the trained text classification model. The technical solution of this application, by introducing adversarial training, enables the text classification model to learn common features of various categories rather than focusing on learning simple features of a particular category. This allows the text classification model to learn deeper representations of categories, improving the generalization and robustness of the text classification model.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Image classification model training method and device, computer equipment and storage medium

    CN111353542A

  • Text classification model training method and device, text classification method and device and storage medium

    CN112131366A