Model training method and related device

Abstract

The embodiment of the invention discloses a model training method in the field of artificial intelligence and a related device, and the method comprises the steps: determining the negative sample confidence degree corresponding to each unlabeled sample in an unlabeled sample set based on a negative sample measurement mechanism; selecting a target negative sample from the unmarked sample set according to the negative sample confidence corresponding to each unmarked sample; constructing a first training sample set based on the target negative sample and the target positive sample; selecting a reference positive sample and a reference negative sample according to the negative sample confidence degrees corresponding to the unmarked samples except the target negative sample in the unmarked sample set; constructing a second training sample set based on the target negative sample, the reference negative sample, the target positive sample and the reference positive sample; and training a target classification model based on the first training sample set and the second training sample set. According to the method, the model obtained by training can more comprehensively and accurately identify various input data.

Description

technical field [0001] The present application relates to the technical field of artificial intelligence, in particular to a model training method and related devices. Background technique [0002] In practical applications, there are many situations where there are only positive samples and a large number of unlabeled samples. This situation usually occurs because it is difficult to obtain negative samples, or the negative samples are too diverse and dynamically changing. [0003] For the above situation, PU learning (Positive and Unlabeled data learning) in semi-supervised learning is usually used to solve the problem. PU learning is usually implemented based on a two-step method; the first step is to use a preset classification method to select negative samples with high confidence from unlabeled samples; the second step is to use the selected negative samples and the correct ones themselves. The positive samples form the training sample set, and then use the training sa...

AI Technical Summary

Problems solved by technology

[0004] However, the inventors of the present application found that the implementation method of the above PU learning has the following defects: in the second step, only the negative samples and positive samples selected by the first step are used for model training, and the rest are not selected as The unlabeled samples of negative samples will lead to limitations of the trained model, and it is difficult to fully and accurately identify various input data; the reason is that unlabeled samples that are not selected as negative samples often account for a large proportion, which contains If there is a wealth of key information, giving up learning this part of key information during model training will make it difficult for the trained model to accurately identify input data that includes such key information

Images