Badcase discovery method and system based on small sample learning

Abstract

The invention discloses a badcase discovery method and system based on small sample learning, and the system applies the method, and the method comprises the steps: carrying out the data preprocessing, randomly obtaining a plurality of small samples from a labeled training corpus, and dividing the samples into a support set and a target set; pre-training the model, inputting the sample into a basic model, combining the basic model with two attention mechanisms, and performing training to obtain a badcase discovery model; predicting data assembly: according to online prediction result semantics, obtaining small samples under the online prediction result semantics from the labeled training corpus, and forming prediction data by the small samples and the sentence to be predicted; and performing data prediction: inputting the prediction data into the badcase discovery model, performing prediction to obtain prediction semantics of a to-be-predicted sentence, comparing the prediction semantics with online prediction result semantics, and judging whether the to-be-predicted sentence is badcase or not. By adopting the method to discover the badcase, quick positioning can be performed in massive data, and error data can be accurately obtained.

Description

technical field [0001] The invention belongs to the technical field of computer data processing, and in particular relates to a badcase discovery method and system based on small sample learning. Background technique [0002] In supervised classification learning, a model with a good offline test set effect often has a certain gap when it is applied online. This requires continuous optimization to address these gaps, so that the online application effect of the model is getting better and better. The performance of these gaps is that when the model predicts and applies online data, the proportion of misclassified data (often referred to as badcase) will be high. The optimization process needs to find these misclassified data, and then analyze the wrong ones in a targeted manner. cause and resolve. The amount of online data is very large, often tens of millions or even billions of visits. If you rely on manpower, it will be very time-consuming and labor-intensive to locate b...

AI Technical Summary

Problems solved by technology

Using random sampling technology (random sample), randomly thousands of online prediction results, and then hand them over to humans for labeling and judgment. The manpower is limited, the random coverage is small, and it cannot cover as many badcases as possible.

Use the threshold selection method of the prediction model (for example, the prediction probability threshold is between 0.5 and 0.7), and manually screen and judge the key points of the data within the threshold. The data source is biased, and the badcase with high prediction probability cannot be found.

Images