Correction model training method, question correction method, device, equipment and medium

By generating sample input text and combining predicted score rate, confidence and true score rate for parameter iteration, the problem of scoring accuracy and flexibility in automatic correction of science answers is solved, and efficient and accurate correction results are achieved.

CN119740207APending Publication Date: 2025-04-01IFLYTEK CO LTD
0 Cites 2 Cited by

Patent Information

Application Number
CN202411941415.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-12-26
Publication Date
2025-04-01

AI Technical Summary

Technical Problem

In the automatic correction of science answers, the scoring accuracy is low and lacks flexibility, especially when adapting to questions with different total scores and accuracy requirements.

Method used

By generating sample input text, using the initial model for prediction, combining the predicted score rate, confidence and true score rate for parameter iteration, to obtain the corrected model. This method transforms the automatic batching task from classification problems to regression problems, and introduces precision selectors and adversarial training techniques.

Benefits of technology

Improve the accuracy and flexibility of correcting answer questions, and can adapt to questions with different total scores, without retraining the model or adjusting the classification standards, improving the model training efficiency and accuracy of scores.

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

Abstract

The present invention relates to the field of artificial intelligence technology, and provides a correction model training method, a question correction method, a device, an equipment and a medium, wherein the training method comprises: generating a sample input text based on a sample question text, a sample answer text and a sample answer text; determining a true score rate based on the total score of the sample question and the true score corresponding to the sample answer; inputting the sample input text into an initial model to obtain a prediction result output by the initial model; and performing parameter iteration on the initial model based on the predicted score rate and confidence in the prediction result and the true score rate to obtain a correction model. The present invention scores by predicting the score rate, can flexibly adapt to questions with different total scores, does not need to retrain the model or adjust the classification standard, improves the model training efficiency, and can improve the accuracy of the correction score while maintaining high efficiency by introducing confidence and precision prediction.
Need to check novelty before this filing date? Find Prior Art