arxiv: 2604.15699 · v1 · submitted 2026-04-17 · 💻 cs.LG · cs.SI

Recognition: unknown

Graph self-supervised learning based on frequency corruption

Guanfeng Liu, Haojie Li, Junwei Du, Mengjiao Zhang, Qiang Hu, Yan Wang

Authors on Pith no claims yet

Pith reviewed 2026-05-10 08:30 UTC · model grok-4.3

classification 💻 cs.LG cs.SI

keywords graphlearningcorruptedfrequencygraphshigh-frequencyself-supervisedfc-gssl

0 comments

The pith

FC-GSSL improves graph SSL by generating high-frequency biased corrupted graphs via low-frequency contribution-based corruption, reconstructing low-frequency features in an autoencoder, and aligning multi-view representations to fuse frequency bands.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Graphs contain signals at different frequencies: low-frequency parts often capture broad structures while high-frequency parts capture local details. Many existing self-supervised methods on graphs tend to over-focus on local patterns and under-use high-frequency signals. FC-GSSL deliberately corrupts the graph by targeting nodes and edges that contribute most to low-frequency content, producing views that are richer in high-frequency information. These corrupted graphs are fed into an autoencoder whose target is to recover the original low-frequency and general features. Multiple sampling strategies create diverse corrupted versions whose intersections and unions are used to generate varied views. The model is trained to make node representations consistent across these views. This forces the learned embeddings to combine information from multiple frequency ranges rather than relying on any single band. The approach is evaluated on node classification, graph-level prediction, and transfer learning tasks across 14 datasets.

Core claim

Experiments on 14 datasets across node classification, graph prediction, and transfer learning show that FC-GSSL consistently improves performance and generalization.

Load-bearing premise

That corrupting nodes and edges according to their low-frequency contributions produces graphs biased toward high-frequency information whose use as autoencoder inputs with low-frequency reconstruction targets will force the model to fuse multi-frequency information and reduce overfitting to local patterns.

Figures

Figures reproduced from arXiv: 2604.15699 by Guanfeng Liu, Haojie Li, Junwei Du, Mengjiao Zhang, Qiang Hu, Yan Wang.

**Figure 1.** Figure 1: An overview of FC-GSSL corresponding rank indices. The specific formula is as follows: Q ∼ 𝑀𝑢𝑙𝑡𝑖𝑛𝑜𝑚𝑖𝑎𝑙 (𝑅𝐼 (𝐶),𝑇 ), (4) where 𝑅𝐼(·) is defined as the ranking and indexing operation, and Q is the sampled item set derived from the rank-based sampling strategy. Corrupted Item Selection Strategy. To mitigate the model’s overreliance on specific high-frequency signals and the bias induced by specific sampling … view at source ↗

**Figure 2.** Figure 2: The Influence of the loss weights 𝛼 and 𝛽. 0.1 0.2 0.3 0.4 0.5 0.6 0.7 The value of rN 38 39 40 41 42 43 Accuracy (%) (a) Actor 0.1 0.2 0.3 0.4 0.5 0.6 0.7 The value of rE 38 39 40 41 42 43 Accuracy (%) (b) Actor 0.1 0.2 0.3 0.4 0.5 0.6 0.7 The value of rN 64 65 66 67 68 69 70 Accuracy (%) (c) molbbbp 0.1 0.2 0.3 0.4 0.5 0.6 0.7 The value of rE 64 65 66 67 68 69 70 Accuracy (%) (d) molbbbp [PITH_FULL_IMAG… view at source ↗

**Figure 3.** Figure 3: The Influence of the sampling rates 𝑟𝑁 and 𝑟𝐸. the additional information provided by the L𝑒𝑑𝑔𝑒 term, which is controlled by 𝛼, is limited. Furthermore, edge corruption helps align representations from different corrupted graphs, enhancing the model’s sensitivity to the L𝑎𝑙𝑖𝑔𝑛 term, which is controlled by 𝛽. (2) As shown in [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

read the original abstract

Graph self-supervised learning can reduce the need for labeled graph data and has been widely used in recommendation, social networks, and other web applications. However, existing methods often underuse high-frequency signals and may overfit to specific local patterns, which limits representation quality and generalization. We propose Frequency-Corrupt Based Graph Self-Supervised Learning (FC-GSSL), a method that builds corrupted graphs biased toward high-frequency information by corrupting nodes and edges according to their low-frequency contributions. These corrupted graphs are used as inputs to an autoencoder, while low-frequency and general features are reconstructed as supervision targets, forcing the model to fuse information from multiple frequency bands. We further design multiple sampling strategies and generate diverse corrupted graphs from the intersections and unions of the sampling results. By aligning node representations from these views, the model can discover useful frequency combinations, reduce reliance on specific high-frequency components, and improve robustness. Experiments on 14 datasets across node classification, graph prediction, and transfer learning show that FC-GSSL consistently improves performance and generalization.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

FC-GSSL adds low-frequency-weighted corruption plus intersection-union sampling to graph SSL and reports gains on 14 datasets, but never checks whether the corruption actually boosts high-frequency content.

read the letter

The main takeaway is that this paper gives graph SSL a concrete frequency-aware corruption rule: nodes and edges are removed or altered in proportion to their low-frequency contributions so the resulting graphs lean toward high-frequency signals. Those corrupted graphs go into an autoencoder whose targets are low-frequency and general features, and the authors add multiple sampling passes whose intersections and unions create extra diverse views that get aligned. That combination is the actual new piece; most prior graph SSL either uses random augmentations or standard contrastive losses without this explicit frequency bias step. The experiments cover 14 datasets across node classification, graph prediction, and transfer learning and claim steady improvements, which is a reasonable empirical footprint for the area. The multi-view sampling strategy is a practical addition that should help robustness. The soft spot is exactly the one the stress test flags: there is no spectral check, such as Laplacian quadratic forms or frequency energy comparisons, showing that the corrupted graphs really do carry more high-frequency content than the originals. Without that measurement the performance lift could come from generic augmentation or the alignment step rather than the claimed multi-frequency fusion. The abstract also gives no ablations or error bars, so the full paper needs to supply those to make the causal story convincing. This work is aimed at people already doing graph SSL for recommendation or social-network tasks who want a drop-in way to push representations past local patterns. A reader who cares about frequency-domain ideas in graphs will find the mechanism worth trying even if the explanatory link stays partly open. It is solid enough to send to peer review; a referee would likely ask for the missing spectral verification and a few targeted ablations but would not dismiss the idea outright.

Referee Report

2 major / 1 minor

Summary. The paper proposes Frequency-Corrupt Based Graph Self-Supervised Learning (FC-GSSL). It constructs corrupted graphs by removing or perturbing nodes and edges weighted by their low-frequency contributions, thereby biasing inputs toward high-frequency signals. These corrupted graphs are fed to an autoencoder whose reconstruction targets are low-frequency and general features; multiple sampling strategies generate diverse views whose node representations are aligned. The method is evaluated on 14 datasets spanning node classification, graph prediction, and transfer learning, with the claim of consistent performance and generalization gains.

Significance. If the frequency-corruption mechanism is shown to reliably increase high-frequency energy and the resulting multi-frequency fusion demonstrably improves representations beyond generic augmentations, the approach could usefully extend graph SSL by mitigating local-pattern overfitting. The multi-view sampling and alignment component is a constructive design choice that could be adopted more broadly.

major comments (2)

[Abstract] Abstract and method description: the central mechanistic claim—that node/edge corruption weighted by low-frequency contributions produces inputs whose spectrum is shifted toward high-frequency content, which the autoencoder then fuses with low-frequency targets—is not directly tested. No spectral analysis (e.g., quadratic form x^T L x or energy in Laplacian eigenvectors) comparing corrupted versus original graphs is reported; without this verification the performance gains on the 14 datasets could arise from generic augmentation rather than the claimed frequency fusion.
[Experiments] Experiments: the manuscript asserts consistent improvement across 14 datasets but supplies neither quantitative tables with per-dataset metrics and baselines, ablation studies isolating the frequency-corruption component, nor error bars or statistical tests. This absence prevents evaluation of effect size, reliability, and whether the gains are attributable to the proposed mechanism.

minor comments (1)

[Method] Notation for frequency contributions and sampling strategies should be defined more explicitly (e.g., precise formulas for low-frequency weighting and intersection/union operations) to allow reproduction.

Circularity Check

0 steps flagged

No circularity: empirical proposal validated on external data

full rationale

The paper defines FC-GSSL as a concrete augmentation procedure (low-frequency-weighted node/edge corruption) whose intended effect on the autoencoder is stated as a design goal rather than derived by equation. No self-definitional loops, fitted parameters renamed as predictions, or load-bearing self-citations appear in the provided text. Performance claims rest on experiments across 14 independent datasets, not on internal algebraic equivalence to the inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based solely on the abstract, the method assumes standard graph signal processing notions of frequency decomposition without introducing new free parameters, axioms beyond domain conventions, or invented entities.

axioms (1)

domain assumption Graph data admits a meaningful decomposition into low- and high-frequency components that can be used to guide corruption.
The corruption rule and reconstruction target both rely on distinguishing frequency contributions.

pith-pipeline@v0.9.0 · 5484 in / 1349 out tokens · 38222 ms · 2026-05-10T08:30:08.283194+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

69 extracted references · 13 canonical work pages · 1 internal anchor

[1]

Deyu Bo, Yuan Fang, Yang Liu, and Chuan Shi. 2023. Graph contrastive learning with stable and scalable spectral encoding.Advances in Neural Information Processing Systems36 (2023), 45516–45532

2023
[2]

Deyu Bo, Chuan Shi, Lele Wang, and Renjie Liao. 2023. Specformer: Spectral graph neural networks meet transformers.arXiv preprint arXiv:2303.01028(2023)

work page arXiv 2023
[3]

Deyu Bo, Xiao Wang, Yang Liu, Yuan Fang, Yawen Li, and Chuan Shi. 2023. A survey on spectral graph neural networks.arXiv preprint arXiv:2302.05631(2023)

work page arXiv 2023
[4]

Pietro Bongini, Monica Bianchini, and Franco Scarselli. 2021. Molecular gen- erative graph neural networks for drug discovery.Neurocomputing450 (2021), 242–252

2021
[5]

Xuheng Cai, Chao Huang, Lianghao Xia, and Xubin Ren. 2023. LightGCL: Simple yet effective graph contrastive learning for recommendation.arXiv preprint arXiv:2302.08191(2023)

work page arXiv 2023
[6]

Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. InICML. 1597–1607

2020
[7]

Vijay Prakash Dwivedi, Chaitanya K Joshi, Anh Tuan Luu, Thomas Laurent, Yoshua Bengio, and Xavier Bresson. 2023. Benchmarking graph neural networks. Journal of Machine Learning Research24, 43 (2023), 1–48

2023
[8]

Vijay Prakash Dwivedi, Anh Tuan Luu, Thomas Laurent, Yoshua Bengio, and Xavier Bresson. 2021. Graph neural networks with learnable structural and positional representations.arXiv preprint arXiv:2110.07875(2021)

work page arXiv 2021
[9]

Moshe Eliasof, Fabrizio Frasca, Beatrice Bevilacqua, Eran Treister, Gal Chechik, and Haggai Maron. 2023. Graph positional encoding via random feature propa- gation. InICML. 9202–9223

2023
[10]

Jingwei Guo, Kaizhu Huang, Xinping Yi, and Rui Zhang. 2023. Graph neural networks with diverse spectral filtering. InWWW. 306–316

2023
[11]

Kaveh Hassani and Amir Hosein Khasahmadi. 2020. Contrastive multi-view representation learning on graphs. InICML. 4116–4126

2020
[12]

Mingguo He, Zhewei Wei, Hongteng Xu, et al. 2021. Bernnet: Learning arbitrary graph spectral filters via bernstein approximation.NeurIPS34 (2021), 14239– 14251

2021
[13]

Zhenyu Hou, Yufei He, Yukuo Cen, Xiao Liu, Yuxiao Dong, Evgeny Kharlamov, and Jie Tang. 2023. Graphmae2: A decoding-enhanced masked self-supervised graph learner. InWWW. 737–746

2023
[14]

Zhenyu Hou, Xiao Liu, Yukuo Cen, Yuxiao Dong, Hongxia Yang, Chunjie Wang, and Jie Tang. 2022. Graphmae: Self-supervised masked graph autoencoders. In KDD. 594–604

2022
[15]

Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. 2020. Open graph benchmark: Datasets for machine learning on graphs.Advances in neural information processing systems 33 (2020), 22118–22133

2020
[16]

Yinan Huang, William Lu, Joshua Robinson, Yu Yang, Muhan Zhang, Stefanie Jegelka, and Pan Li. 2023. On the stability of expressive positional encodings for graphs.arXiv preprint arXiv:2310.02579(2023)

work page arXiv 2023
[17]

Dejun Jiang, Zhenxing Wu, Chang-Yu Hsieh, Guangyong Chen, Ben Liao, Zhe Wang, Chao Shen, Dongsheng Cao, Jian Wu, and Tingjun Hou. 2021. Could graph neural networks learn better molecular representation for drug discovery? A comparison study of descriptor-based and graph-based models.Journal of cheminformatics13, 1 (2021), 12

2021
[18]

Chanwoo Kim, Jinkyu Sung, Yebonn Han, and Joonseok Lee. 2025. Graph Spectral Filtering with Chebyshev Interpolation for Recommendation. InSIGIR. 1964– 1974

2025
[19]

Thomas N Kipf and Max Welling. 2016. Variational graph auto-encoders.arXiv preprint arXiv:1611.07308(2016)

work page arXiv 2016
[20]

Yunwen Lei, Tianbao Yang, Yiming Ying, and Ding-Xuan Zhou. 2023. General- ization analysis for contrastive representation learning. InICML. 19200–19227

2023
[21]

Haojie Li, Junwei Du, Guanfeng Liu, Feng Jiang, Yan Wang, and Xiaofang Zhou
[22]

Intent Propagation Contrastive Collaborative Filtering.IEEE Transactions on Knowledge and Data Engineering01 (2025), 1–14

2025
[23]

Haojie Li, Guanfeng Liu, Qiang Hu, Yan Wang, Dunwei Gong, and Junwei Du
[24]

Candidate-aware graph prompt-tuning for recommendation.Pattern Recognition(2025), 111733

2025
[25]

Jing Li. 2024. Area under the ROC Curve has the most consistent evaluation for binary classification.PloS one19, 12 (2024), e0316019

2024
[26]

Jialu Li, Yu Wang, Pengfei Zhu, Wanyu Lin, and Qinghua Hu. 2024. What matters in graph class incremental learning? An information preservation perspective. NeurIPS37 (2024), 26195–26223

2024
[27]

Jintang Li, Ruofan Wu, Wangbin Sun, Liang Chen, Sheng Tian, Liang Zhu, Changhua Meng, Zibin Zheng, and Weiqiang Wang. 2023. What’s behind the mask: Understanding masked graph modeling for graph autoencoders. InKDD. 1268–1279

2023
[28]

Zihan Lin, Changxin Tian, Yupeng Hou, and Wayne Xin Zhao. 2022. Improving graph collaborative filtering with neighborhood-enriched contrastive learning. InWWW. 2320–2329

2022
[29]

Chuang Liu, Yuyao Wang, Yibing Zhan, Xueqi Ma, Dapeng Tao, Jia Wu, and Wenbin Hu. 2024. Where to mask: structure-guided masking for graph masked autoencoders. InIJCAI. 2180–2188

2024
[30]

Jiahao Liu, Dongsheng Li, Hansu Gu, Tun Lu, Peng Zhang, Li Shang, and Ning Gu. 2023. Personalized graph signal processing for collaborative filtering. In WWW. 1264–1272

2023
[31]

Nian Liu, Xiao Wang, Deyu Bo, Chuan Shi, and Jian Pei. 2022. Revisiting graph contrastive learning from the perspective of graph spectrum.NeurIPS35 (2022), 2972–2983

2022
[32]

Yang Liu, Deyu Bo, Wenxuan Cao, Yuan Fang, Yawen Li, and Chuan Shi. 2025. Graph Positional Autoencoders as Self-supervised Learners. InKDD. 1867–1878

2025
[33]

Yanbei Liu, Yu Zhao, Xiao Wang, Lei Geng, and Zhitao Xiao. 2024. Multi-scale subgraph contrastive learning.arXiv preprint arXiv:2403.02719(2024)

work page arXiv 2024
[34]

Zhiyuan Liu, Yaorui Shi, An Zhang, Enzhi Zhang, Kenji Kawaguchi, Xiang Wang, and Tat-Seng Chua. 2023. Rethinking tokenizer and decoder in masked graph modeling for molecules.NeurIPS, 25854–25875

2023
[35]

Zaiqiao Meng, Shangsong Liang, Hongyan Bao, and Xiangliang Zhang. 2019. Co-embedding attributed networks. InWSDM. 393–401

2019
[36]

Viet Anh Nguyen, Nhat Khang Ngo, and Truong Son Hy. 2024. Range-aware Positional Encoding via High-order Pretraining: Theory and Practice.arXiv preprint arXiv:2409.19117(2024)

work page arXiv 2024
[37]

Soon Hyeok Park and Kyoungok Kim. 2023. Collaborative filtering recommenda- tion system based on improved Jaccard similarity.Journal of Ambient Intelligence and Humanized Computing14, 8 (2023), 11319–11336

2023
[38]

Hongbin Pei, Bingzhe Wei, Kevin Chen-Chuan Chang, Yu Lei, and Bo Yang
[39]

Geom-gcn: Geometric graph convolutional networks.arXiv preprint arXiv:2002.05287(2020)

work page arXiv 2002
[40]

Yifang Qin, Wei Ju, Yiyang Gu, Ziyue Qiao, Zhiping Xiao, and Ming Zhang. 2025. Polycf: Towards optimal spectral graph filters for collaborative filtering.ACM Transactions on Information Systems43, 4 (2025), 1–28

2025
[41]

Amin Salehi and Hasan Davulcu. 2019. Graph attention auto-encoders.arXiv preprint arXiv:1905.10715(2019)

work page arXiv 2019
[42]

Yifei Shen, Yongji Wu, Yao Zhang, Caihua Shan, Jun Zhang, B Khaled Letaief, and Dongsheng Li. 2021. How powerful is graph convolution for recommendation?. InCIKM. 1619–1629

2021
[43]

Shashank Sheshar Singh, Samya Muhuri, Shivansh Mishra, Divya Srivastava, Harish Kumar Shakya, and Neeraj Kumar. 2024. Social network analysis: A survey on process, tools, and application.ACM computing surveys56, 8 (2024), 1–39

2024
[44]

Teague Sterling and John J Irwin. 2015. ZINC 15–ligand discovery for everyone. Journal of chemical information and modeling55, 11 (2015), 2324–2337

2015
[45]

Fan-Yun Sun, Jordan Hoffmann, Vikas Verma, and Jian Tang. 2019. Infograph: Un- supervised and semi-supervised graph-level representation learning via mutual information maximization.arXiv preprint arXiv:1908.01000(2019)

work page arXiv 2019
[46]

Qiaoyu Tan, Ninghao Liu, Xiao Huang, Soo-Hyun Choi, Li Li, Rui Chen, and Xia Hu. 2023. S2gae: Self-supervised graph autoencoders are generalizable learners with graph masking. InWSDM. 787–795

2023
[47]

Shantanu Thakoor, Corentin Tallec, Mohammad Gheshlaghi Azar, Rémi Munos, Petar Veličković, and Michal Valko. 2021. Bootstrapped representation learning on graphs. InICLR

2021
[48]

Yijun Tian, Kaiwen Dong, Chunhui Zhang, Chuxu Zhang, and Nitesh V Chawla
[49]

Heterogeneous graph masked autoencoders. InAAAI. 9997–10005
[50]

Amanda L Traud, Peter J Mucha, and Mason A Porter. 2012. Social structure of facebook networks.Physica A: Statistical Mechanics and its Applications391, 16 (2012), 4165–4180

2012
[51]

Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua Bengio. 2017. Graph attention networks.arXiv preprint arXiv:1710.10903(2017)

work page internal anchor Pith review arXiv 2017
[52]

Petar Veličković, William Fedus, William L Hamilton, Pietro Liò, Yoshua Bengio, and R Devon Hjelm. 2018. Deep graph infomax.arXiv preprint arXiv:1809.10341 (2018)

work page arXiv 2018
[53]

Haorui Wang, Haoteng Yin, Muhan Zhang, and Pan Li. [n. d.]. Equivariant and Stable Positional Encoding for More Powerful Graph Neural Networks. InICLR
[54]

Liang Wang, Xiang Tao, Qiang Liu, and Shu Wu. 2024. Rethinking graph masked autoencoders through alignment and uniformity. InAAAI. 15528–15536

2024
[55]

Wenguan Wang, Yi Yang, and Fei Wu. 2024. Towards data-and knowledge-driven AI: a survey on neuro-symbolic computing.IEEE Transactions on Pattern Analysis and Machine Intelligence(2024)

2024
[56]

Xiyuan Wang and Muhan Zhang. 2022. How powerful are spectral graph neural networks. InICML. 23341–23362

2022
[57]

Zhibiao Wang, Xiao Wang, Haoyue Deng, Nian Liu, Shirui Pan, and Chunming Hu. 2024. Uncovering the redundancy in graph self-supervised learning models. InNeurIPS. 98625–98644

2024
[58]

Jiancan Wu, Xiang Wang, Fuli Feng, Xiangnan He, Liang Chen, Jianxun Lian, and Xing Xie. 2021. Self-supervised graph learning for recommendation. InSIGIR. 726–735

2021
[59]

Zhenqin Wu, Bharath Ramsundar, Evan N Feinberg, Joseph Gomes, Caleb Ge- niesse, Aneesh S Pappu, Karl Leswing, and Vijay Pande. 2018. MoleculeNet: a WWW ’26, April 13–17, 2026, Dubai, United Arab Emirates Haojie Li et al. benchmark for molecular machine learning.Chemical science9, 2 (2018), 513–530

2018
[60]

Jun Xia, Chengshuai Zhao, Bozhen Hu, Zhangyang Gao, Cheng Tan, Yue Liu, Siyuan Li, and Stan Z Li. 2023. Mole-bert: Rethinking pre-training graph neural networks for molecules. (2023)

2023
[61]

Rongwei Xu, Guanfeng Liu, Yan Wang, Xuyun Zhang, Kai Zheng, and Xiaofang Zhou. 2024. Adaptive hypergraph network for trust prediction. InICDE. 2986– 2999

2024
[62]

Zhirui Yang, Yulan Hu, Sheng Ouyang, Jingyu Liu, Shuqiang Wang, Xibo Ma, Wenhan Wang, Hanjing Su, and Yong Liu. 2024. WaveNet: tackling non-stationary graph signals via graph spectral wavelets. InAAAI. 9287–9295

2024
[63]

Yihang Yin, Qingzhong Wang, Siyu Huang, Haoyi Xiong, and Xiang Zhang. 2022. Autogcl: Automated graph contrastive learning via learnable view generators. In AAAI. 8892–8900

2022
[64]

Yuning You, Tianlong Chen, Yang Shen, and Zhangyang Wang. 2021. Graph contrastive learning automated. InICML. 12121–12132

2021
[65]

Yuning You, Tianlong Chen, Yongduo Sui, Ting Chen, Zhangyang Wang, and Yang Shen. 2020. Graph contrastive learning with augmentations.Advances in neural information processing systems33 (2020), 5812–5823

2020
[66]

Hengrui Zhang, Qitian Wu, Junchi Yan, David Wipf, and Philip S Yu. 2021. From canonical correlation analysis to self-supervised graph neural networks.Advances in Neural Information Processing Systems34 (2021), 76–89

2021
[67]

Yige Zhao, Jianxiang Yu, Yao Cheng, Chengcheng Yu, Yiding Liu, Xiang Li, and Shuaiqiang Wang. 2025. Variational Graph Autoencoder for Heterogeneous Information Networks with Missing and Inaccurate Attributes. InKDD. 2067– 2078

2025
[68]

Guanghui Zhu, Wang Lu, Chunfeng Yuan, and Yihua Huang. 2023. Adamcl: Adaptive fusion multi-view contrastive learning for collaborative filtering. In SIGIR. 1076–1085

2023
[69]

Ziyun Zou, Yinghui Jiang, Lian Shen, Juan Liu, and Xiangrong Liu. 2025. Loha: Direct graph spectral contrastive learning between low-pass and high-pass views. InAAAI. 13492–13500. A Model Details A.1 Training Procedure of FC-GSSL In this section, we will illustrate the training process of the FC-GSSL in detail. The pseudo-code of FC-GSSL is shown in Algor...

2025