Multi-View Clustering Based on Adaptive Similarity Graph Joint Optimization

Compared to single-view learning， multi-view learning can often obtain more comprehensive information about the learning object. Therefore， in the field of unsupervised learning， multi-view clustering has received great attention from researchers. Among them，graph based multi-view clustering has made great research progress in recent years. Graph based multi-view clustering generally involves learning similar graphs from the raw data of each view，and then fusing similar graphs between views to obtain the final clustering result. Therefore， the effectiveness of multi-view clustering is determined by the quality of similar graphs and the fusion method of similar graphs. However， existing graph based multi-view clustering methods almost all focus on the fusion of similar graphs between views，and lack attention to the quality of similar graphs themselves. Most of these methods learn similar graphs in isolation from the raw data of each view， and keep the similar graphs unchanged in the subsequent graph fusion process. The similarity graph obtained in this way inevitably contains noise and redundant information，which in turn affects subsequent graph fusion and clustering. The existing a small amount of studies that consider the quality of similarity graphs either directly iterate the construction of similarity graphs and graph fusion processes， or use rank constraints to further initializes in advance during the predefined similarity graph process， or utilizes some underlying structures of similarity graphs to obtain the fused graph. These methods have very little improvement in the quality of similarity graph， so the final clustering performance improvement is also very limited. At the same time，the existing graph based multi-view clustering process lacks comprehensive consideration of consistency and inconsistency between views，which will also seriously affect the final multi-view clustering performance. In order to avoid the adverse effects of low-quality predefined similarity graphs on clustering results， and to comprehensively consider the consistency and inconsistency between views to improve the final clustering effect， this paper proposes a multi-view clustering method based on adaptive similarity graph joint optimization （AJO-MVC）. Firstly， the Hadamard product is used to obtain high-quality consistency information between views，and then the predefined similarity graphs of each view are compared with this information to reconstruct the preset similarity graphs for each view. This process strengthens the consistency between different views and weakens the inconsistency. Secondly， a joint iterative optimization framework for similar graph reconstruction and graph fusion is designed to achieve adaptive improvement of similar graphs， ultimately achieving a joint improvement of similar graphs and clustering results. This AJO-MVC method combines the process of improving similar graphs with the process of graph fusion for adaptive iterative optimization， and continuously strengthens the consistency between views and weakens the inconsistency between views during the iterative optimization. In addition， this AJO-MVC method proposed in this paper also integrates some advantages of existing multi-view clustering methods， such as self-weighting and no need for additional clustering steps. The effectiveness and superiority of our AJO-MVC method have been fully validated through experiments on nine benchmark datasets with eight comparison methods. © 2024 Science Press. All rights reserved.