Design of Multilabel Classification for Unstructured Data Based on Multi-View Orthogonal Basis Extraction

In multi-view multi-label learning, an object can extract its features from multiple different feature extraction methods, so as to obtain multiple views that describing the same object, and transfer the unstructured data such as text data to the structured data. All multiple views from the same object correspond to the same label. Each view plays a different role in multi-label learning, and each view may contain the same and different information at the same time. In this paper, a base-coordinate decomposition method is proposed to reconstruct each view, in which each view has its corresponding base matrix, and all views of each sample share the same coordinates under their own base matrix. This coordinate is the re expression of the original object and contains the feature information of multiple views at the same time. Finally, the coordinate is used as the input of the classifier to obtain the forecast label. Furthermore, we make orthogonal constraints on the base matrix to reduce the redundant information of the base matrix of each view as much as possible and ensure the uniqueness of the internal information of the base matrix. Experiments show the excellent performance of the proposed classifier.