Accurate Graph-Based PU Learning without Class Prior

Yoo, Junehee; Kim, Jung-Hun; Yoon, Hoyoung; Kim, Geonsoo; Jang, Changwon; Kang, U

doi:10.1109/icdm51629.2021.00094

Cited by 5 publications

(5 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In an epidemic network [11], neighbors of positive (affected) nodes are also more likely to be positive. Nevertheless, such information has been ignored by previous works on PU node classification [17,19], which simply use the general PU loss for node classification.…”

Section: Methodsmentioning

confidence: 99%

“…As can be seen, using the PU loss improves the naive GCN baseline much, and Dist-PU significantly outperforms nnPU in most cases. Methods specific to PU node classification (LSDAN [17] and GRAB [19]) generally have better performances than Dist-PU, and their performances are comparable to each other. The proposed PU-GNN achieves the best performance for almost all data sets and label ratios, demonstrating that graph structure plays a critical role in the loss design for PU node classification.…”

Section: Node Classification Performancementioning

confidence: 96%

“…We conducted experiments on four node classification data sets: Cora, Citeseer, Pubmed and DBLP, which are popularly used in literature [8,15,17,19]. The data set statistics are summarized in Table 1.…”

Section: Experiments 41 Experimental Setupmentioning

confidence: 99%

“…The proposed method is compared with the following baselines: (i) naive GCN, which considers all unlabeled nodes as negative; (ii) GCN with nnPU loss [9]; (iii) GCN with Dist-PU loss [20]; (iv) LSDAN [17], which introduces long-range dependency with the nnPU loss [9]; and (v) GRAB [19], which models the given graph as a Markov network and uses loopy belief propagation (LBP) to estimate the label distribution of unlabeled nodes. As in [8,9,20], we use macro F1 score (%) on the testing nodes as performance measure.…”

Section: Experiments 41 Experimental Setupmentioning

confidence: 99%

See 3 more Smart Citations

Positive-Unlabeled Node Classification with Structure-aware Graph Learning

Yang,

Zhang,

Yao

et al. 2023

Proceedings of the 32nd ACM International Conference on Information and Knowledge Management

View full text Add to dashboard Cite

Node classification on graphs is an important research problem with many applications. Real-world graph data sets may not be balanced and accurate as assumed by most existing works. A challenging setting is positive-unlabeled (PU) node classification, where labeled nodes are restricted to positive nodes. It has diverse applications, e.g., pandemic prediction or network anomaly detection. Existing works on PU node classification overlook information in the graph structure, which can be critical. In this paper, we propose to better utilize graph structure for PU node classification. We first propose a distance-aware PU loss that uses homophily in graphs to introduce more accurate supervision. We also propose a regularizer to align the model with graph structure. Theoretical analysis shows that minimizing the proposed loss also leads to minimizing the expected loss with both positive and negative labels. Extensive empirical evaluation on diverse graph data sets demonstrates its superior performance over existing state-of-the-art methods. CCS CONCEPTS• Computing methodologies → Semi-supervised learning settings.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Node Classification Performancementioning

confidence: 96%

Section: Experiments 41 Experimental Setupmentioning

confidence: 99%

Section: Experiments 41 Experimental Setupmentioning

confidence: 99%

See 2 more Smart Citations

Positive-Unlabeled Node Classification with Structure-aware Graph Learning

Yang,

Zhang,

Yao

et al. 2023

Proceedings of the 32nd ACM International Conference on Information and Knowledge Management

View full text Add to dashboard Cite

show abstract

“…Compared to node classification [16] that focuses on the properties of individual nodes, graph classification aims at extracting meaningful substructures of nodes to make a better representation of a graph. A traditional way to solve the problem is to utilize kernel methods [24,26,29,42], while graph neural networks [13,16,17,41,45,47] have shown a better performance with advanced pooling methods [1,21,44] to aggregate node embeddings. We use graph classification as a downstream task for evaluating augmented graphs, since No augmentation SplitOnly MergeOnly NodeSamBase NodeSam (proposed) SubMixBase SubMix (proposed) accurate classification requires separating the core information of each graph from random and unimportant components.…”

Section: Related Workmentioning

confidence: 99%

Model-Agnostic Augmentation for Accurate Graph Classification

Yoo¹,

Shim²,

Kang³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

Given a graph dataset, how can we augment it for accurate graph classification? Graph augmentation is an essential strategy to improve the performance of graph-based tasks, and has been widely utilized for analyzing web and social graphs. However, previous works for graph augmentation either a) involve the target model in the process of augmentation, losing the generalizability to other tasks, or b) rely on simple heuristics that lead to unreliable results. In this work, we introduce five desired properties for effective augmentation. Then, we propose NodeSam (Node Split and Merge) and SubMix (Subgraph Mix), two model-agnostic algorithms for graph augmentation that satisfy all desired properties with different motivations. NodeSam makes a balanced change of the graph structure to minimize the risk of semantic change, while SubMix mixes random subgraphs of multiple graphs to create rich soft labels combining the evidence for different classes. Our experiments on social networks and molecular graphs show that NodeSam and SubMix outperform existing approaches in graph classification.

show abstract