圖對比表徵學習為近年最重要的表徵學習技術之一，其能有效學習節點間的結構與特徵資訊，進而提升模型效能。然而，由於現實世界中所收集到的圖數據常常面臨資料品質不良的問題，意即圖數據發生極端噪聲資訊或極端弱資訊，前者可能包含錯誤或經惡意攻擊後的資訊，後者則反映資訊過於稀缺、資訊量不足，導致模型學習難度提升。此外，由於噪聲資訊與弱資訊的問題常常同時發生且相互影響，如果僅針對單一問題各自處理而非將他們一併考量並進行修正，則模型表徵提取的效果很難提升，進而嚴重影響下游任務的表現。
因此，本研究提出了可自我修正圖對比表徵學習(SC-GCL)，同步處理噪聲資訊與弱資訊。SC-GCL包含兩個模組：雙重模式圖去噪(DMGD)與時序感知對比式學習(TACL)，前者目標是處理圖中噪聲結構與特徵，首先利用特徵的相似性來修補結構，接著再藉由去噪後的結構進行特徵重建，形成結構與特徵彼此協作、強化的機制，提升圖資訊的正確性與多元性；後者目的在於處理弱資訊的問題，進一步強化節點表徵對語意的學習能力。最後在訓練過程中進行迭代優化，實現噪聲資訊與弱資訊的聯合處理。藉由結合DMGD與TACL，SC-GCL能自我修正噪聲資訊與弱資訊，學習穩健且具辨識力的節點表徵，提升下游任務表現。
經實驗驗證，SC-GCL相較於現有主流方法均展現出顯著的性能優勢。首先，當噪聲與弱資訊程度同時設定成0.3、0.4或0.5時，SC-GCL在Cora資料集的準確度各自至少提升了24.7%、23.2%及17.3%，而在CiteSeer與PubMed資料集上也能全面優於現有方法。此外，當噪聲與弱資訊程度進一步設定成多種不同比例的組合時，SC-GCL依舊維持優於其他方法的表現，足以顯示SC-GCL具備極高的環境可適性。由此可見，本論文提出了一個可以妥善處理極端噪聲與極端弱資訊問題的圖表徵學習機制，讓機器學習的各種下游任務在面臨惡劣資訊環境的情況下仍能提供讓人滿意的服務。

Graph contrastive representation learning has recently become an important technique for learning structural and feature information among nodes, thereby enhancing model performance. However, graph data collected from real-world scenarios often suffers from poor data quality, characterized by the presence of either extremely noisy information or extremely weak information. The former may include incorrect or adversarially manipulated data, while the latter reflects a severe lack of information, making representation learning more difficult. Moreover, noisy and weak information frequently coexist and interact with each other. If these issues are addressed independently rather than jointly, the effectiveness of representation learning remains limited, which in turn severely impacts the performance of downstream tasks.
Therefore, this research proposes a Self-Corrective Graph Contrastive Learning (SC-GCL), designed to address noisy and weak information together in graph data. SC-GCL consists of two core modules: Dual-Mode Graph Denoising (DMGD) and Temporal-Aware Contrastive Learning (TACL). The DMGD module aims to denoise both structural and feature information by first rewriting the graph structure based on feature similarity and subsequently reconstructing node features using the denoised structure. This mutually reinforcing process raises the accuracy and diversity of the graph information. The TACL module aims to address weak information by improving the semantic representation of nodes by incorporating temporal awareness into the contrastive learning. An iterative optimization process is applied during training to jointly handle noisy and weak information. By combining DMGD and TACL, SC-GCL can self-correct noisy and weak information, learn robust and discriminative node representations, and improve the performance of downstream tasks.
Experimental results demonstrate that SC-GCL exhibits significant performance advantages over existing mainstream methods. Specifically, when both the noisy and weak information ratios are set to 0.3, 0.4, or 0.5, SC-GCL achieves accuracy improvements of at least 24.7%, 23.2%, and 17.3%, respectively, on the Cora dataset. It also consistently outperforms existing methods on the CiteSeer and PubMed datasets. Furthermore, even when the noise and weak information levels are configured with various combinations of different ratios, SC-GCL maintains superior performance, indicating its strong adaptability to diverse environments. These findings confirm that the proposed framework offers a robust graph representation learning capable of effectively handling extremely noisy and weak information. As a result, it enables downstream machine learning tasks to maintain satisfactory performance even under adverse information conditions.

[1] N. A. Asif et al., "Graph neural network: A comprehensive review on non-euclidean space," Ieee Access, vol. 9, pp. 60588-60606, 2021.
[2] U. A. Bhatti, H. Tang, G. Wu, S. Marjan, and A. Hussain, "Deep learning with graph convolutional networks: An overview and latest applications in computational intelligence," International Journal of Intelligent Systems, vol. 2023, no. 1, p. 8342104, 2023.
[3] F. Chen, Y.-C. Wang, B. Wang, and C.-C. J. Kuo, "Graph representation learning: a survey," APSIPA Transactions on Signal and Information Processing, vol. 9, p. e15, 2020.
[4] L.-H. Chen et al., "Erase: Error-resilient representation learning on graphs for label noise tolerance," in Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, 2024, pp. 270-280.
[5] Y. Chen, L. Wu, and M. Zaki, "Iterative deep graph learning for graph neural networks: Better and robust node embeddings," Advances in neural information processing systems, vol. 33, pp. 19314-19326, 2020.
[6] K. W. Church, "Word2Vec," Natural Language Engineering, vol. 23, no. 1, pp. 155-162, 2017.
[7] E. Dai, C. Aggarwal, and S. Wang, "Nrgnn: Learning a label noise resistant graph neural network on sparsely and noisily labeled graphs," in Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining, 2021, pp. 227-236.
[8] E. Dai, W. Jin, H. Liu, and S. Wang, "Towards robust graph neural networks for noisy graphs with sparse labels," in Proceedings of the fifteenth ACM international conference on web search and data mining, 2022, pp. 181-191.
[9] K. Ding, X. Ma, Y. Liu, and S. Pan, "Divide and Denoise: Empowering Simple Models for Robust Semi-Supervised Node Classification against Label Noise," in Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024, pp. 574-584.
[10] J. Feng, Y. Chen, F. Li, A. Sarkar, and M. Zhang, "How powerful are k-hop message passing graph neural networks," Advances in Neural Information Processing Systems, vol. 35, pp. 4776-4790, 2022.
[11] W. Feng et al., "Graph random neural networks for semi-supervised learning on graphs," Advances in neural information processing systems, vol. 33, pp. 22092-22103, 2020.
[12] S. Geisler, T. Schmidt, H. Şirin, D. Zügner, A. Bojchevski, and S. Günnemann, "Robustness of graph neural networks at scale," Advances in Neural Information Processing Systems, vol. 34, pp. 7637-7649, 2021.
[13] A. Goknil et al., "A systematic review of data quality in CPS and IoT for industry 4.0," ACM Computing Surveys, vol. 55, no. 14s, pp. 1-38, 2023.
[14] M. Gori, G. Monfardini, and F. Scarselli, "A new model for learning in graph domains," in Proceedings. 2005 IEEE international joint conference on neural networks, 2005., 2005, vol. 2: IEEE, pp. 729-734.
[15] A. Grover and J. Leskovec, "node2vec: Scalable feature learning for networks," in Proceedings of the 22nd ACM SIGKDD international conference on Knowledge discovery and data mining, 2016, pp. 855-864.
[16] W. Jin, T. Derr, Y. Wang, Y. Ma, Z. Liu, and J. Tang, "Node similarity preserving graph convolutional networks," in Proceedings of the 14th ACM international conference on web search and data mining, 2021, pp. 148-156.
[17] W. Jin, Y. Ma, X. Liu, X. Tang, S. Wang, and J. Tang, "Graph structure learning for robust graph neural networks," in Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining, 2020, pp. 66-74.
[18] T. N. Kipf and M. Welling, "Semi-supervised classification with graph convolutional networks," arXiv preprint arXiv:1609.02907, 2016.
[19] X. Liu et al., "Graph neural networks with adaptive residual," Advances in Neural Information Processing Systems, vol. 34, pp. 9720-9733, 2021.
[20] X. Liu et al., "Elastic graph neural networks," in International Conference on Machine Learning, 2021: PMLR, pp. 6837-6849.
[21] X. Liu et al., "Self-supervised learning: Generative or contrastive," IEEE transactions on knowledge and data engineering, vol. 35, no. 1, pp. 857-876, 2021.
[22] Y. Liu, K. Ding, J. Wang, V. Lee, H. Liu, and S. Pan, "Learning strong graph neural networks with weak information," in Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2023, pp. 1559-1571.
[23] D. Luo et al., "Learning to drop: Robust graph neural network via topological denoising," in Proceedings of the 14th ACM international conference on web search and data mining, 2021, pp. 779-787.
[24] W. Parker. "Zillow’s Shuttered Home-Flipping Business Lost $881 Million in 2021." https://www.wsj.com/business/earnings/zillows-shuttered-home-flipping-business-lost-881-million-in-2021-11644529656 (accessed.
[25] B. Perozzi, R. Al-Rfou, and S. Skiena, "Deepwalk: Online learning of social representations," in Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining, 2014, pp. 701-710.
[26] T. C. Redman. "Seizing Opportunity in Data Quality." https://sloanreview.mit.edu/article/seizing-opportunity-in-data-quality/ (accessed.
[27] F. P. Releases. "Poor-Quality Data Imposes Costs and Risks on Businesses, Says New Forbes Insights Report." https://www.forbes.com/sites/forbespr/2017/05/31/poor-quality-data-imposes-costs-and-risks-on-businesses-says-new-forbes-insights-report/ (accessed.
[28] S. Sanyal and P. Zhang, "Improving quality of data: IoT data aggregation using device to device communications," IEEE Access, vol. 6, pp. 67830-67840, 2018.
[29] P. Sen, G. Namata, M. Bilgic, L. Getoor, B. Galligher, and T. Eliassi-Rad, "Collective classification in network data," AI magazine, vol. 29, no. 3, pp. 93-93, 2008.
[30] D. Shah, J. Wang, and Q. P. He, "Feature engineering in big data analytics for IoT-enabled smart manufacturing–Comparison between deep learning and statistical learning," Computers & Chemical Engineering, vol. 141, p. 106970, 2020.
[31] I. Spinelli, S. Scardapane, and A. Uncini, "Missing data imputation with adversarially-trained graph convolutional networks," Neural Networks, vol. 129, pp. 249-260, 2020.
[32] K. Sun, Z. Lin, and Z. Zhu, "Multi-stage self-supervised learning for graph convolutional networks on graphs with few labeled nodes," in Proceedings of the AAAI conference on artificial intelligence, 2020, vol. 34, no. 04, pp. 5892-5899.
[33] Y. Sun, S. Wang, X. Tang, T.-Y. Hsieh, and V. Honavar, "Adversarial attacks on graph neural networks via node injections: A hierarchical reinforcement learning approach," in Proceedings of the Web Conference 2020, 2020, pp. 673-683.
[34] H. Taguchi, X. Liu, and T. Murata, "Graph convolutional networks for graphs containing missing features," Future Generation Computer Systems, vol. 117, pp. 155-168, 2021.
[35] S. Wan, Y. Zhan, L. Liu, B. Yu, S. Pan, and C. Gong, "Contrastive graph poisson networks: Semi-supervised learning with extremely limited labels," Advances in Neural Information Processing Systems, vol. 34, pp. 6316-6327, 2021.
[36] T. Yang et al., "You Can't Ignore Either: Unifying Structure and Feature Denoising for Robust Graph Learning," in Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, 2024, pp. 4178-4182.
[37] J. Zhao, Q. Wen, M. Ju, C. Zhang, and Y. Ye, "Self-supervised graph structure refinement for graph neural networks," in Proceedings of the sixteenth ACM international conference on web search and data mining, 2023, pp. 159-167.
[38] B. Zhou et al., "Robust graph representation learning for local corruption recovery," in Proceedings of the ACM Web Conference 2023, 2023, pp. 438-448.