Missing label multi-label classification method based on example-level and label-level association

Abstract

The invention requests to protect a missing label multi-label classification method based on example-level and label-level association, and the method comprises the steps: S1, inputting a feature matrix of a training sample, constructing a feature-based sample neighbor graph through a linear recombination strategy, mining the geometric structure information of the sample, and obtaining an example-level association matrix; S2, inputting a label matrix of a training sample, and constructing a label-based semantic association graph through a low-rank representation method to mine semantic association information of labels to obtain a label-level association matrix; S3, utilizing Laplace manifold regularization to associate and construct the two labels into two regularization items; and S4, constructing an objective function and solving the objective function. According to the multi-label classification method, the relevance of the example level and the label level is combined, and the classification effect of the multi-label classification method can be effectively improved under the condition that the labels are partially lost.

Description

technical field [0001] The invention belongs to the technical field of machine learning, and in particular relates to a multi-label classification method under the condition of missing labels. Background technique [0002] Multi-label learning is widely used in tasks such as text classification and image annotation. Unlike single-label learning, each example in multi-label learning usually has multiple labels simultaneously. The correlation between labels, including instance-level correlation and label-level correlation, plays a key role in improving the performance of multi-label learning algorithms. [0003] Although most of the existing multi-label learning algorithms have achieved certain results in dealing with multi-label learning problems, most of them assume that the label set of training samples is complete. Due to the difficulty of manual labeling and huge time overhead, in practical applications, some labels of training samples are often missing, and this missin...

AI Technical Summary

Problems solved by technology

However, when mining example-level associations, these methods usually use the Gaussian function and the k-nearest neighbor strategy to calculate the example similarity matrix, so that only the similarity between pairs of examples can be calculated, and the σ value in the Gaussian function is difficult to determine

In addition, when mining tag-level associations, some methods use cosine similarity to calculate the tag similarity matrix, so that the calculated similarity in the absence of tags is usually inaccurate, but will have a negative impact on the results of the algorithm

Images