Method and system for assisting in labeling model training data

Abstract

The invention discloses a method and a system for assisting in labeling model training data. The method comprises the following steps: S1, constructing two training sets; s2, training two classifiers which are in one-to-one correspondence with the training sets, respectively predicting categories of training data in the training sets of the opposite side through the two classifiers, obtaining error data of which prediction results are inconsistent with categories labeled by operators, and extracting error features from the error data; s3, updating the to-be-labeled data pool, obtaining a plurality of new training data based on the new to-be-labeled data pool, and distributing the new training data to the two training sets; and S4, repeating the steps S2-S3 based on the two new training sets until the total number of the training data in the two training sets reaches a preset value, and exporting all the training data in the two training sets. The system for assisting in labeling the model training data adopts the method, some valuable model training data are screened out, and the labeling efficiency and labeling quality are improved in the labeling process of the model training data at the same time.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to a method and system for assisting in labeling model training data. Background technique [0002] Benefiting from the rapid development of computing resources and the massive data brought by the popularization of the Internet, artificial intelligence technology has developed rapidly. Among them, supervised learning plays an important role. Supervised learning relies on a large amount of labeled data, and the efficiency of data labeling is low. These two characteristics have long hindered the development and application of AI technology. Although we can invest more human resources to alleviate this problem, the labeling of data by a large number of ordinary people with different identities, different ideas and no professional training is bound to give a negative result to the labeling results (that is, some of the above Labeled data) brings a lot of "noise...

AI Technical Summary

Problems solved by technology

Although we can invest more human resources to alleviate this problem, the labeling of data by a large number of ordinary people with different identities, different ideas and no professional training is bound to give a negative result to the labeling results (that is, some of the above Labeled data) brings a lot of "noise", and these "noises" will have a non-negligible impact on the training process of the subsequent trained algorithm model

Images