Research map / Graph learning / Graph self-supervised learning

Global Graph Propagation with Hierarchical Information Transfer for Incomplete Contrastive Multi-view Clustering

AAAI 2025 · Graph learning › Graph self-supervised learning · approach: Multi-view graph clustering · 13 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes GHICMC, an end-to-end incomplete multi-view clustering framework that uses view-specific GCN autoencoders, layer-wise global graph propagation across views to impute missing representations, and a weight-shared pseudo-classifier trained with contrastive clustering.

Problem

Incomplete multi-view clustering, where samples are missing some views, and existing methods fail to mine information in missing data and separate representation learning from clustering.

Approach

Global graph propagation passes each GCN layer's output of the consensus graph into the same-layer GCNs of every view, so missing features are imputed using both intra-view and cross-view relations, and representation learning and contrastive clustering are optimized jointly end-to-end.

Key results

• On MSRC-V1 at missing rate 0.5, GHICMC reaches 91.14 ACC, 82.70 NMI, 80.61 ARI, versus the second-best PIMVC/DCP-level baselines (79.14 ACC for PIMVC).
• On 100leaves at missing rate 0.5, GHICMC reaches 70.46 ACC vs 61.60 for PIMVC.
• Ablation on MSRC-V1 (missing rate 0.5) shows removing the contrastive clustering module drops ACC from 91.14 to 67.62.

Measured effects reported in the paper

Surprising findings

• Removing the contrastive clustering loss causes the largest performance drop in the ablation, more than removing the graph propagation module.

Limitations

• Evaluated only on five small-to-medium benchmark datasets.
• Hyperparameters (alpha, beta, K) are sensitive in some ranges, e.g. poor when alpha much larger than beta.
• Reported results on several datasets remain low in absolute terms (e.g. LandUse-21 ACC around 25-28%).

Experimental setup

Datasets: HandWritten MSRC-V1 Scene-15 LandUse-21 100leaves

Benchmarks: HandWritten MSRC-V1 Scene-15 LandUse-21 100leaves

Metrics: ACC NMI ARI

Baselines: BSV Concat DAIMC UEAF DCP ANIMC LSIMVC PIMVC

Builds on: graph convolutional network contrastive clustering t-distribution KL self-training (DEC-style)

Closest papers by meaning

• Incomplete Contrastive Multi-View Clustering with High-Confidence Guiding (AAAI 2024 · similarity 0.91)
• Multi-View Graph Clustering via Node-Guided Contrastive Encoding (ICML 2025 · similarity 0.90)
• Medusa: A Multi-Scale High-order Contrastive Dual-Diffusion Approach for Multi-View Clustering (CVPR 2025 · similarity 0.89)
• SURER: Structure-Adaptive Unified Graph Neural Network for Multi-View Clustering (AAAI 2024 · similarity 0.89)
• Deep Contrastive Graph Learning with Clustering-Oriented Guidance (AAAI 2024 · similarity 0.89)
• Adaptive Feature Imputation with Latent Graph for Deep Incomplete Multi-View Clustering (AAAI 2024 · similarity 0.88)