Multi-view semi-supervised learning for classification on dynamic networks

In recent decades, the task of graph-based multi-view learning has become a fundamental research problem, which could integrate data from multiple sources to improve performance. The dynamic networks could be treated as one kind of multi-view network, but it is continually evolving and leads to entirely different observations at multiple epochs. In this paper, we treat these observations as multiple views and seek a semi-supervised multi-view approach to address the classification problem. Therefore, we propose Multi-view Semi-supervised learning for Classification on Dynamic networks (MSCD). With the aid of total variation regularization, MSCD can obtain a sparse and smooth combination of the views and a better classification result. From the theoretical point of view, the MSCD model is decomposed into simpler sub-problems, which can be effectively solved under the Alternating Direction Method of Multipliers (ADMM) framework. Extensive experiments on both synthetic and real-world datasets show that our model can outperform the state-of-the-art approaches. (C) 2020 Elsevier B.V. All rights reserved.