Sample generation method and device, model training method and system, and storage medium

By automatically generating question-and-answer sample data, using the target word replacement method to generate adversarial attack samples and train the question-and-answer model, the problems of low efficiency and manual dependence in existing technologies are solved, and the robustness and predictive ability of the question-and-answer system are improved.

CN116108154BActive Publication Date: 2026-07-21JD DIGITS HAIYI INFORMATION TECHNOLOGY CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
JD DIGITS HAIYI INFORMATION TECHNOLOGY CO LTD
Filing Date
2023-02-08
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

In existing technologies, the generation of adversarial attack samples in question-answering systems is inefficient and requires manual intervention, making them unsuitable for large-scale use and unable to effectively improve the robustness and predictive ability of the model.

Method used

By automatically generating question-and-answer sample data, using the target word replacement method, determining the attack text based on the perturbation capability of the replaced text, generating adversarial attack samples, and training the question-and-answer model with these samples, the robustness and attack capability of the model are improved.

Benefits of technology

It improves the efficiency of generating adversarial attack samples and the attack capability of the question-answering model, enhances the robustness and predictive ability of the model, and reduces manual costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116108154B_ABST
    Figure CN116108154B_ABST
Patent Text Reader

Abstract

The disclosure provides a sample generation method and device, a model training method and system and a storage medium, and relates to the technical field of machine learning. The sample generation method provided by the disclosure comprises the following steps: determining a target word in a first text according to first sample data, wherein the first sample data comprises the first text, a question and an answer; replacing the target word in a predetermined area of the first text with a corresponding replacement word, and determining an attack text according to the perturbation ability of a model after the replacement; and obtaining adversarial attack sample data according to the attack text, the question and the answer. Through the method, the attack ability and efficiency of the question and answer model are improved while the efficiency of generating the adversarial attack sample is improved, so as to help improve the prediction ability of the question and answer model and improve the accuracy of the question and answer system.
Need to check novelty before this filing date? Find Prior Art