A Semi-supervised Multi-View Clustering Method Based on Structural Constraints

Abstract

The invention discloses a semi-supervised multi-view clustering method based on structural constraints. The method optimizes the structural information of multi-view data through regression, that is, the normalized category matrix, and uses prior information to directly guide the multi-view data structure. Information learning to further impose structural constraints on the normalized category matrix, so that the performance of multi-view clustering can be effectively improved when prior information is given. Considering that different perspectives play different roles in learning, the present invention embeds feature learning into the learning of multi-view data structure information to further improve clustering.

Description

technical field [0001] The invention relates to the technical field of pattern recognition, in particular to a multi-view data clustering method based on semi-supervised conditions. Background technique [0002] With the rapid development of computer vision technology and multimedia technology, data often show different feature description methods. For example, a single webpage can be described by information such as pictures, text, and hyperlinks; pictures can be described by different visual description operators (such as SIFT and GIST features) for encoding. The above data is called multi-view data, and each view corresponds to a feature set. The explosive growth of multi-view data has promoted the development of multi-view learning and produced a wide range of applications. Multi-view clustering, as a basic task of multi-view learning, aims to mine the complementary characteristics of information between different views to improve data clustering performance. [0003]...

AI Technical Summary

Problems solved by technology

Traditional semi-supervised multi-view clustering methods generally use soft regularization (soft regularization) or hard constraints (hard constraints) to use semi-supervised information, but they cannot explicitly use the structural information shown by prior information to Multi-view clustering is used for guidance, and at the same time, it cannot handle the selection of multi-view data features well.

Therefore, the traditional semi-supervised multi-view clustering method cannot effectively improve the clustering performance and cannot effectively meet the needs of semi-supervised multi-view clustering.

Images