Nonnegative spectral embedding learning with adaptive neighbors for multi-view clustering.

Abstract

Graph-based multi-view clustering (MVC) methods often rely on fixed or manually constructed similarity graphs and involve multiple sensitive hyperparameters, which limits their robustness and practical applicability. To address these issues, we propose Nonnegative Spectral Embedding with Adaptive Neighbors (NSEAN), a unified one-stage MVC framework that integrates per-view adaptive graph learning with nonnegative spectral embedding. NSEAN jointly learns adaptive similarity graphs and a consensus spectral embedding that directly serves as the clustering indicator matrix, thereby eliminating the need for post-processing. By enforcing nonnegativity and orthogonality, the learned embedding admits a clear and interpretable cluster-assignment structure. To efficiently optimize the coupled constraints, an Augmented Lagrangian Multiplier (ALM) strategy is employed to ensure stable and effective optimization. Extensive experiments on real-world multi-view datasets demonstrate that NSEAN consistently achieves competitive or superior clustering performance while requiring only a single hyperparameter number of neighbors k, to which the model is empirically insensitive, thus avoiding cumbersome parameter tuning. The code is available at https://github.com/haha1206/NSEAN.