RTGD-MVC: Robust Tensor Learning with Graph Diffusion for Scalable Multi-view Graph Clustering

This repository provides the official MATLAB implementation for the paper "RTGD-MVC: Robust Tensor Learning with Graph Diffusion for Scalable Multi-view Graph Clustering".

We are thrilled to announce that our paper, "RTGD-MVC: Robust Tensor Learning with Graph Diffusion for Scalable Multi-view Graph Clustering", has been accepted by ACM Multimedia 2025. It has also been recommended for an Oral Presentation, a testament to its strong peer-review evaluation.

Key Features:
🔹 Robust Tensor Learning with Graph Diffusion for Scalable Multi-view Graph Clustering.
🔹 One-click reproducible experiments with hyperparameter optimization
🔹 Comprehensive baseline comparisons (see baseline/ directory)
🔹 Supports common datasets (BBC, BDGP, CCV, etc.).

1. Requirements
2. Project Structure
3. Quick Start
4. Dataset Preparation
5. Output Results
6. Parameters
7. License

• MATLAB ≥ R2019b
• MATLAB Toolboxes:
  • Statistics and Machine Learning Toolbox
  • (Optional) Parallel Computing Toolbox (for large datasets)

RTGD-MVC/
├── data/                   # Dataset storage
│   └── BBC.mat             # Sample dataset file
├── exp/                    # Experiment scripts
│   ├── run_demo.m          # Main experiment script
│   └── result_RTGD-MVC/    # Results storage (auto-generated)
├── lib/                    # Utility functions
│   └── NormalizeFea.m      # Data normalization
├── utils/                  # Core algorithm implementation
│   ├── Construct_FB.m      # Anchor graph construction
│   └──RTGD.m              # Main algorithm function
├── baseline/               # Baseline implementations
│   ├── AWMVC/              # Adaptive Weighted MVC
│   ├── FPMVS-CAG/          # Fast Probabilistic MVC
│   └── ...                 # Other baselines
└── docs/                   # Supplementary materials

The code is publicly available at 'https://anonymous.4open.science/r/RTGD-MVC-6646/'

1. Launch MATLAB and navigate to the exp folder:

   cd /path/to/RTGD-MVC/exp

2. Execute the demo script:

   run run_demo.m

3. Results will be saved in exp/result_RTGD-MVC/.

1. Place Your Dataset

   • Save your dataset as a .mat file in the data/ folder.
   • Example: For dataset mydata, save it as data/mydata.mat.

2. Dataset Format Requirements
   Ensure your .mat file contains:

   % Variables:
   % - X: Cell array of multi-view data {nView × 1}, each view is [nSmp × nFeature]
   % - Y: Ground truth labels [nSmp × 1]
   load('mydata.mat'); 

🔄 Example Workflow

1. Add mydata.mat to data/:

   RTGD-MVC/
   └── data/
       └── mydata.mat
   

2. Set dataset = 'mydata'; in run_demo.m
3. Run the code. Results will use your custom dataset.

The experimental results will be saved in the following directory structure:

exp/result_RTGD-MVC/
└── {dataset_name}/               # e.g., BBC/
    ├── {dataset_name}_RTGD.mat   # Aggregated results (best parameters)
    └── {dataset_name}_RTGD_param{1-N}.mat   # Per-parameter results

D:\Sean\MVC\RTGD-MVC\exp\result_RTGD-MVC\BBC\
├── BBC_RTGD.mat                 # Best results across all parameters
├── BBC_RTGD_param1.mat          # Results for parameter set 1
├── BBC_RTGD_param2.mat          # Results for parameter set 2
└── ...                          # Additional parameter results

Key parameters in run_demo.m:

% Hyperparameter grid search ranges:
lambda_s = 10.^(-6:1:0);  % Sparse error weight (1e-6 to 1)
delta_s = 10.^(-6:1:0);   % Convergence threshold (1e-6 to 1)
nAnch_s = nClus.*(2:1:8); % Anchors: 2×nClus to 8×nClus
ks_s = [10];               % k-Nearest Neighbors
eta_s = [1];               % Graph regularization

This project is licensed under the MIT License. See LICENSE for details.