README.md

ADGC: Awesome Deep Graph Clustering

ADGC is a collection of state-of-the-art (SOTA), novel deep graph clustering methods (papers, codes and datasets). Any other interesting papers and codes are welcome. Any problems, please contact [email protected]. If you find this repository useful to your research or work, it is really appreciate to star this repository. ❤️

---

What's Deep Graph Clustering?

Deep graph clustering, which aims to reveal the underlying graph structure and divide the nodes into different groups, has attracted intensive attention in recent years.

​

Important Survey Papers

Papers

1. K-Means: "Algorithm AS 136: A k-means clustering algorithm" [paper|code]
2. DEC (ICML16): "Unsupervised Deep Embedding for Clustering Analysis" [paper|code]
3. DCN (ICML17): "Towards k-means-friendly spaces: Simultaneous deep learning and clustering" [paper|code]
4. IDEC (IJCAI17): "Improved Deep Embedded Clustering with Local Structure Preservation" [paper|code]
5. GAE/VGAE : "Variational Graph Auto-Encoders" [paper|code]
6. DAEGC (IJCAI19): "Attributed Graph Clustering: A Deep Attentional Embedding Approach" [paper|code]
7. ARGA/ARVGA (TCYB19): "Learning Graph Embedding with Adversarial Training Methods" [paper|code]
8. MCGC (TIP19): "Multiview Consensus Graph Clustering" [paper|code]
9. SDCN/SDCN_Q (WWW20): "Structural Deep Clustering Network" [paper|code]
10. AGE (SIGKDD20): "Adaptive Graph Encoder for Attributed Graph Embedding" [paper|code]
11. MVGRL (ICML20): "Contrastive Multi-View Representation Learning on Graphs" [paper|code]
12. DFCN (AAAI21): "Deep Fusion Clustering Network" [paper|code]
13. MCGC (NIPS21): "Multi-view Contrastive Graph Clustering" [paper|code]
14. GCC (ICCV21): "Graph Contrastive Clustering" [paper|code]
15. DCRN (AAAI22): "Deep Graph Clustering via Dual Correlation Reduction" [paper|code]

Benchmark Datasets

We divide the datasets into two categories, i.e. graph datasets and non-graph datasets. Graph datasets are some graphs in real-world, such as citation networks, social networks and so on. Non-graph datasets are NOT graph type. However, if necessary, we could construct "adjacency matrices" by K-Nearest Neighbors (KNN) algorithm.

Quick Start

• Step1: Download all datasets from [Google Drive] | [Baidu Netdisk, code: 1234]. Optionally, download some of them from URLs in the tables (Google Drive)
• Step2: Unzip them to ./dataset/
• Step3: Change the type and the name of the dataset in main.py
• Step4: Run the main.py

Code

• utils.py
  1. load_graph_data: load graph datasets
  2. load_data: load non-graph datasets
  3. normalize_adj: normalize the adjacency matrix
  4. diffusion_adj: calculate the graph diffusion
  5. construct_graph: construct the knn graph for non-graph datasets
  6. numpy_to_torch: convert numpy to torch
  7. torch_to_numpy: convert torch to numpy
• clustering.py
  1. setup_seed: fix the random seed
  2. evaluation: evaluate the performance of clustering
  3. k_means: K-means algorithm
• visualization.py
  1. t_sne: t-SNE algorithm
  2. similarity_plot: visualize cosine similarity matrix of the embedding or feature

Datasets Details

1. Graph Datasets

   Dataset	Samples	Dimension	Edges	Classes	URL
   CORA	2708	1433	6632	7	cora.zip
   CITESEER	3327	3703	6215	6	citeseer.zip
   PUBMED	19717	500	44325	3	pubmed.zip
   DBLP	4057	334	3528	4	dblp.zip
   CITE	3327	3703	4552	6	cite.zip
   ACM	3025	1870	13128	3	acm.zip
   AMAP	7650	745	119081	8	amap.zip
   AMAC	13752	767	245861	10	amac.zip
   CORAFULL	19793	8710	63421	70	corafull.zip
   COCS

2. Non-graph Datasets

   Dataset	Samples	Dimension	Type	Classes	URL
   USPS	9298	256	Image	10	usps.zip
   HHAR	10299	561	Record	6	hhar.zip
   REUT	10000	2000	Text	4	reut.zip

Citation

If you use code or datasets in this repository for your research, please cite our paper.

@inproceedings{
}