Multi-label classification method based on gravity model

Abstract

The invention relates to the field of machine learning, and especially relates to a multi-label classification method based on a gravity model, comprising the steps of: acquiring a sample set with labels to serve as a training sample set; calculating the distances from one training sample to other training samples and performing sorting to obtain a near neighbor set of the training sample; in thenear neighbor set, building a positive correlation matrix based on positive correlation properties among the labels, and building a negative correlation matrix based on negative correlation propertiesamong the labels; calculating a near neighbor set of a to-be-detected sample, and building a to-be-detected positive correlation matrix and a to-be-detected negative correlation matrix according to the near neighbor set; obtaining a positive correlation data granule and a negative correlation data granule based on the to-be-detected positive correlation matrix and the to-be-detected negative correlation matrix; and building the gravity model, and performing classification through a gravity relationship between the to-be-detected sample and the positive correlation data granule as well as thenegative correlation data granule. The multi-label classification method of the invention has the beneficial effects that: consideration about a negative correlation relationship between the labels isintroduced, the correlation between the labels is taken full advantage of, and the correlation relationship is mined in the near neighbor set, thereby avoiding global computing and reducing complexity.

Description

technical field [0001] The invention relates to the field of machine learning, in particular to a multi-label classification method based on a gravity model. Background technique [0002] In the field of machine learning, classification problems occupy a large proportion. Traditional machine learning is mainly based on two-class classification or multi-class classification, and its purpose is to accurately classify each data to be classified into a certain class. Such single-classification problems and multi-classification problems may be collectively referred to as single-label classification. In practical applications, most classification tasks need to face the multi-label classification problem. For example, a picture, the content of the picture may contain various elements, such as the beach, the sea, high-rise buildings, characters, etc. Classifying such images is a multi-label classification task. [0003] The existing multi-label classification methods mainly adop...

AI Technical Summary

Problems solved by technology

However, the existing methods cannot make good use of label correlation or lead to a sharp increase in the complexity of classification methods in the exploration of label correlation.

Images