Abstract
<title>Abstract</title> <p>Existing multi-view spectral clustering methods often suffer from unstable cluster assignments, high computational complexity, and insufficient exploitation of high-order correlations across multiple views. To address these challenges, this paper proposes a fast one-step multi-view clustering method based on tensor log-determinant regularization. Specifically, spectral clustering and nonnegative matrix factorization are unified within a single optimization framework, where a high-quality consensus nonnegative embedding matrix is learned through an adaptive weighting strategy. Furthermore, tensor log-determinant regularization is imposed on the multi-view spectral embedding matrices to perform low-rank modeling, providing a tighter approximation to the underlying tensor rank while more effectively capturing high-order cross-view correlations. Experimental results on ten real-world benchmark datasets demonstrate that the proposed method consistently outperforms representative state-of-the-art methods in terms of both clustering performance and scalability.</p>