Low-Rank Tensor Regularized Graph Fuzzy Learning for Multi-View Data Processing

Multi-view data processing is an effective tool to differentiate the levels of consumers on electronics. Recently, the graph based multi-view clustering methods have attracted widespread attention because they can obtain the relationships of multi-view data points efficiently. However, there exist several shortcomings on most existing graph based clustering methods. Firstly, the mostly adopted Euclidean distance can not extract the nonlinear manifold structure. Secondly, graph based methods are mainly hard clustering methods, which means that each data point belongs to only the one cluster exactly. Thirdly, the high-dimension information between multiple views are not taken into account. Thus, a low-rank tensor regularized graph fuzzy learning (LRTGFL) method for multi-view data processing is proposed. In LRTGFL, Jensen-Shannon divergence is adopted to replace the Euclidean distance for obtaining more completely nonlinear structures. In addition, fuzzy learning is adopted to make graph clustering be a soft clustering method. Furthermore, a tensor nuclear norm based on the tensor singular value decomposition (t-SVD) is adopted to take advantage of the high-dimension information. Then, alternating direction method of multipliers (ADMM) is adopted to solve the LRTGFL model. Finally, the effectiveness and superiority of LRTGFL are demonstrated by comparing with various state-of-the-art algorithms on eight real-world datasets.