Multi-view multi-mark classification method based on view category characteristic learning

Abstract

The invention relates to a multi-mark learning technology in the field of machine learning, and relates to a multi-view multi-mark classification method based on view category characteristic learning.The method comprises the following steps: S1, acquiring training data, and establishing a class mark matrix; S2, constructing a linear model of mapping the visual angle characteristic data after thecategory marking to a category marking matrix; S3, on the basis of the linear model, establishing contribution degree models of all visual angle characteristics; S4, adopting a regular item to constrain the contribution degree model of the visual angle characteristics, and enabling each visual angle characteristic data to have consistency on a prediction result; S5, adopting manifold regularization to constrain the similarity of the model coefficients corresponding to the related category marks; S6, performing mark prediction, giving a test sample t, and substituting the test sample t into thesteps S1 to S2; and S5, obtaining a fusion prediction value. According to the technical scheme provided by the invention, multi-source information is effectively utilized to learn the discriminationperformance of different features on the category mark in each view angle, and a multi-mark learning task is better carried out.

Description

Technical field [0001] The present invention relates to a multi-label learning technology in the field of machine learning, relates to a perspective generic feature learning and classification technology in multi-perspective multi-label learning, and particularly relates to a multi-perspective multi-label classification method based on perspective generic feature learning. Background technique [0002] In a big data environment, the semantics and knowledge of data are often expressed through content information in multiple modalities or perspectives, and each data sample may belong to multiple semantic tags at the same time. For example, in a text classification task, a document may contain multiple data types such as text, images, videos, and hyperlinks, and contain multiple semantic topics at the same time, such as "machine learning", "data mining" and "multi-label learning" Wait. Multi-view and multi-label learning is an important research direction in the field of data minin...

AI Technical Summary

Problems solved by technology

Although the existing method considers the contribution of a single view feature data, due to the noise and redundant features in the multi-view data, the directly learned view contribution weight may be inaccurate.

And the existing methods do not consider the contribution of each feature in the single-view data to the category label

Images