Semi-supervised classification algorithm based on dimension weighting and view angle feature consistency

Abstract

The invention provides a semi-supervised classification algorithm based on dimension weighting and view angle feature consistency. The method comprises the following steps: firstly, constructing a similarity matrix of each view angle of multi-view-angle data by adopting a self-adaptive local structure learning method; then, taking an average value of similarity matrixes of all view angles as an initial consistency similarity matrix, and constructing a multi-view-angle semi-supervised classification model based on dimension weighting and view angle feature consistency; then, solving the model by adopting an alternate iterative updating method until a final label matrix is obtained; and finally, obtaining labels of the samples according to the label matrix, and completing sample classification. According to the classification model constructed by the invention, the construction similarity matrix is combined with the label deduction, so that the influence of the composition quality on theclassification result is reduced; and since the feature dimensions in the view angle are weighted and the local structure information of the data is considered, a better classification result can beobtained.

Description

technical field [0001] The invention belongs to the technical field of machine learning and data mining, and in particular relates to a semi-supervised classification algorithm based on dimension weighting and visual angle feature consistency. Background technique [0002] With the advent of the era of big data, information in many real scenes can be obtained through different channels, different angles, different modes, and different features. Faced with these multi-source data, how to efficiently and accurately fuse these information through a certain strategy to complete specific tasks and apply them in actual scenarios has important research significance. [0003] Assuming that there is "complementarity" and "consistency" in multi-view data sets, multi-view learning refers to a method that describes the research object from multiple perspectives, and then integrates information from multiple perspectives for learning. Semi-supervised classification refers to using a sma...

AI Technical Summary

Problems solved by technology

However, these regression-based ideas above only consider the linear relationship between samples and labels, and are not satisfied with nonlinear relationships.

In graph-based semi-supervised classification, the quality of the constructed similarity graph will greatly affect the final classification results, and since the construction of the similarity graph and label inference are considered as two separate steps, the relationship between the two is ignored

In addition, these methods only consider the differences in features between views, and ignore the differences between dimensions within views, thereby ignoring the local structure information of the data.

Therefore, the classification accuracy of these methods suffers

Images