Multi-View Graph Clustering via Node-Guided Contrastive Encoding
Authors: Yazhou Ren, Junlong Ke, Zichen Wen, Tianyi Wu, Yang Yang, Xiaorong Pu, Lifang He
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate its robust performance on six homophilic and heterophilic multi-view benchmark datasets. We conduct extensive experiments on six homophilic and heterophilic benchmark datasets to evaluate the performance of NGCE. Our results show that the proposed NGCE adeptly accommodates both homophilic and heterophilic datasets within the multi-view graph clustering domain, achieving state-of-the-art performance metrics. |
| Researcher Affiliation | Academia | 1School of Computer Science and Engineering, University of Electronic Science and Technology of China 2Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China 3Department of Computer Science and Engineering, Lehigh University. Correspondence to: Junlong Ke <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 Multi-View Graph Clustering via Node Guided Contrastive Encoding (NGCE) Input: The adjacency matrices {An}V n=1 and node feature matrices {Xn}V n=1 Initialization phase Apply the k-means algorithm to original node feature matrices {Xn}V n=1 for the pseudo label matrix Y of the first iteration Train phase while not reaching the maximum iterations or not reaching exit condition do for n in 1, 2, ..., V do Calculate Zn and Xn pred based on Eq. (3) and (4) Calculate Sn based on Eq. (5) If it is the first iteration do Calculate ωn based on Eq. (6) with pseudo label matrix YT from Initialization phase Calculate ωn based on Eq. (6) with pseudo label matrix Y from last iteration Calculate b An and e Xn based on Eq. (7) and (9) Calculate X n by X n = GCN( b A, e X) based on Eq. (10) Calculate hn and ωn h based on Eq. (13) and (16) end for Update H based on Eq. (15) for v in 1, 2, ..., V do Update Y with H base on k-means end for end while Output: Consensus embedding H |
| Open Source Code | No | The paper does not provide an explicit statement about the release of source code or a link to a code repository. |
| Open Datasets | Yes | We used six MVG datasets. The homophilous graph datasets include: ACM (Fan et al., 2020), DBLP (Fan et al., 2020) and IMDB (Fan et al., 2020). The heterophilous graph datasets include: Texas, Chameleon (Rozemberczki et al., 2021) and Wisconsin (Pei et al., 2020). In the appendix, we provide details and sources of the datasets. ... 1https://dl.acm.org/ 2https://dblp.uni-trier.de/ 3https://www.imdb.com/ 4http://www.cs.cmu.edu/afs/cs.cmu.edu/project/theo11/www/wwkb 5https://github.com/benedekrozemberczki/MUSAE/ 6http://www.cs.cmu.edu/afs/cs.cmu.edu/project/theo11/www/wwkb |
| Dataset Splits | No | The paper uses various datasets for clustering tasks but does not specify any training/test/validation dataset splits, percentages, or detailed splitting methodologies. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU models, CPU types) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific details about software dependencies, such as library names with version numbers. |
| Experiment Setup | Yes | Nknn indicates the number of nearest neighbors selected, which is fixed at 20 in our implementation. ... Herein, t denotes the current training epoch and T1 is a predefined hyperparameter; this loss component is thereby activated only when t T1. ... The depth of our GCN is defined in terms of order, which is set as a fixed hyperparameter in our framework. ... The hyperparameter ρ facilitates the modulation of view weights, either enhancing or diminishing their impact. ... For the sake of cross-dataset robustness and simplicity in hyperparameter tuning, we omitted the weight of the loss term. |