Extended category type data quick labeling method

Abstract

The invention relates to an extended category type data rapid labeling method. The method comprises the following steps of (1) obtaining an initial data set with a unified label A; (2) further dividing A into a plurality of subclasses such as a, b, c and the like in combination with the existing data set; (3) selecting a part of the initial data set as a sample, manually modifying the annotation file, and quickly renaming the annotation as a, b, c and the like from A to obtain an annotation sample; (4) based on the annotation sample, training a model by using a neural network, and pre-annotating residual data in the initial data set by using the obtained model to obtain pre-annotated data; (5) performing manual fine adjustment on the pre-annotated data to ensure that all related data labels are accurate, performing model training optimization based on the initial data and the fine adjustment data, training a better model based on more annotation samples, and finally accurately annotating all the data to improve the annotation efficiency; the method can meet the requirements of quick and accurate labeling, and better serves the practical application.

Description

Technical field [0001] The present invention belongs to the field of computer technology, and more specifically relates to a method for rapid labeling of expanded category data. Background technique [0002] In recent years, the attention of artificial intelligence technology has continued to increase, and the collection of a large number of social capital, intelligence, and data resources has driven the continuous development of artificial intelligence technology research. The classic technology of artificial intelligence is machine learning technology, and the emergence of deep learning technology enables many applications of machine learning, expanding the scope of artificial intelligence, and its foundation is to add annotation information to a large amount of data. Fast and accurate completion of data labeling is a key step in the realization of topics and projects. [0003] At present, the common data labeling method is manual labeling with the help of data labeling tools, w...

AI Technical Summary

Problems solved by technology

The method can label the original data set, which is more efficient than manual labeling, but still has limitations such as relying on labeling tools and high labeling costs

Images