Structure-aware Hybrid-order Similarity Learning for Multi-view Unsupervised Feature Selection

TL;DR

SHINE-FS tackles multi-view unsupervised feature selection by adaptively learning consensus anchors and an anchor graph to capture cross-view relationships, then constructing a second-order graph from anchor–sample relations. It further combines first- and second-order similarities into a hybrid-order graph to preserve both local and global data structures, guiding discriminative feature selection through an iterative, efficient optimization scheme. Extensive experiments on eight real-world datasets demonstrate significant improvements over state-of-the-art methods in ACC and NMI, with ablations confirming the value of each component. The approach offers a robust, scalable solution for MUFS and lays groundwork for potential semi-supervised extensions.

Abstract

Multi-view unsupervised feature selection (MUFS) has recently emerged as an effective dimensionality reduction method for unlabeled multi-view data. However, most existing methods mainly use first-order similarity graphs to preserve local structure, often overlooking the global structure that can be captured by second-order similarity. In addition, a few MUFS methods leverage predefined second-order similarity graphs, making them vulnerable to noise and outliers and resulting in suboptimal feature selection performance. In this paper, we propose a novel MUFS method, termed Structure-aware Hybrid-order sImilarity learNing for multi-viEw unsupervised Feature Selection (SHINE-FS), to address the aforementioned problem. SHINE-FS first learns consensus anchors and the corresponding anchor graph to capture the cross-view relationships between the anchors and the samples. Based on the acquired cross-view consensus information, it generates low-dimensional representations of the samples, which facilitate the reconstruction of multi-view data by identifying discriminative features. Subsequently, it employs the anchor-sample relationships to learn a second-order similarity graph. Furthermore, by jointly learning first-order and second-order similarity graphs, SHINE-FS constructs a hybrid-order similarity graph that captures both local and global structures, thereby revealing the intrinsic data structure to enhance feature selection. Comprehensive experimental results on real multi-view datasets show that SHINE-FS outperforms the state-of-the-art methods.

Figures (6)

• Figure 1: The framework of the proposed SHINE-FS.
• Figure 2: ACC of different methods on eight datasets with different feature selection ratios.
• Figure 4: Visualizations of the similarity graphs learned by different graph-based methods on MSRA dataset.
• Figure 5: The t-SNE visualizations on Mfeat dataset.
• Figure 6: ACC of SHINE-FS with varying parameters $\gamma$, $\beta$, $\eta$ and feature selection ratios on MSRA dataset.