Semi-supervised text classification model training method, text classification method, system, device and medium

Abstract

The invention provides a semi-supervised text classification model training method and system, a text classification method and system, equipment and a medium. The training method comprises the stepsof acquiring an initial sample set; enhancing the unlabeled sample to obtain a data enhanced sample; inputting the unlabeled sample and the data enhancement sample into a text classification model toobtain an embedded vector and a prediction probability belonging to each classification label; for each unlabeled sample, obtaining an embedded vector mean value of the unlabeled sample and the corresponding data enhancement sample as a new sample; for each unlabeled sample, obtaining a label estimation result of a new sample after sharpening the prediction probability mean value of the unlabeledsample and the corresponding data enhancement sample belonging to each classification label; verifying whether the new sample is credible or not, and if yes, marking the new sample as a credible new sample; constructing a confidence sample set according to the labeled sample and the trusted new sample, and expanding to obtain a target sample set; and training the text classification model according to the target sample set. According to the method, the text classification accuracy is improved under the condition of lack of annotation samples.

Description

technical field [0001] The present invention relates to the field of deep learning, in particular to a semi-supervised text classification model training method, text classification method, system, equipment and media. Background technique [0002] Machine learning methods attempt to use historical data of tasks to improve the performance of tasks. In order to obtain good learning performance, machine learning methods, such as supervised learning methods, usually require historical data to be clearly labeled (called labeled data) and require a large amount of labeled data. However, in many real-world tasks, because the acquisition of labeled data requires a lot of human and material resources, labeled data is usually scarce, while a large amount of unlabeled historical data (called unlabeled data) can be easily obtained. How to use a large amount of unlabeled data to help improve the performance obtained by using only a small amount of labeled data has become an important t...

AI Technical Summary

Problems solved by technology

[0004] However, the above two semi-supervised methods are dedicated to image data processing and cannot be directly used for text classification to improve the accuracy of text classification in the absence of labeled samples

In addition, the above two semi-supervised methods do not consider the model's confidence in the predicted value of unlabeled samples, and may use unlabeled samples with unreliable low-entropy labels for sample expansion, affecting the accuracy of the training model

Images