Incomplete multi-view clustering via attention-based contrast learning

Multi-view clustering (MVC) is an essential and challenging task in machine learning and data mining. In recent years, this field has attracted a lot of attention and achieved remarkable results. The success of multi-view clustering relies heavily on the consistency and integrity of data views to ensure complete data information. In the process of data collection and transmission, data inevitably be lost, which leads to the occurrence of partial view unalignment (VN) and partial view missing (VM). This situation reduces the available information and increases the difficulty of joint learning of multi-view data. To address the incomplete information problem, in this article, we present a novel incomplete multi-view clustering via attention-based contrast learning framework (MCAC) to address the VN and VM puzzles. Due to the diversity of different views, negative samples are formed by randomly selecting some cross-view samples from positive samples, then computing the correlation between local features and latent features for each view by maximizing mutual information and, fusing each specific low-dimensional representation into a joint representation through an attention fusion layer, in addition, adding noise contrast loss reduces or even eliminates the effect of negative samples. MCAC conducts experiments on seven multi-view datasets and demonstrates the effectiveness compared to eleven state-of-the-art methods on the multi-view clustering task.