Abstract:Graph neural networks (GNNs) are effective for structured data analysis but face reduced learning accuracy due to noisy connections and the necessity for explicit graph structures and labels. This requirement constrains their usability in diverse graph-based applications. In order to address these issues, considerable research has been directed toward graph structure learning that aims to denoise graph structures concurrently and refine GNN parameters. However, existing graph structure learning approaches encounter several challenges, including dependence on label information, underperformance of learning algorithms, insufficient data augmentation methods, and limitations in performing downstream tasks. We propose Uogtag, an unsupervised graph structure learning framework to address these challenges. Uogtag optimizes graph topology through the selection of suitable graph learners for the input data and incorporates contrastive learning with adaptive data augmentation, enhancing the learning and applicability of graph structures for downstream tasks. Comprehensive experiments on various real-world datasets demonstrate Uogtag's efficacy in managing noisy graphs and label scarcity.

What problem does this paper attempt to address?

### Problems the paper attempts to solve This paper aims to solve the challenges faced by graph neural networks (GNNs) when dealing with graph data with noisy connections and scarce labels. Specifically, the paper proposes an unsupervised graph structure learning framework named Uogtag to address the following issues: 1. **Dependence on label information**: Traditional graph structure learning methods (such as supervised GSL) rely heavily on manually labeled labels to enhance the structure, which is often impractical in real - world applications because the cost of data labeling is high. 2. **Limitations of graph learners**: Existing graph structure learning methods usually use a single graph learner, which may lead to significant performance differences on different datasets. 3. **Insufficient data augmentation methods**: Current graph learning strategies lack effective data augmentation methods, limiting the potential for structure discovery and optimization. 4. **Task - specific limitations**: Methods focused on specific tasks (such as node classification) may lead to overly specialized structures, thereby limiting their wide application in other tasks (such as link prediction or clustering). ### Solutions To address the above problems, the paper proposes Uogtag, an unsupervised graph structure learning framework. The main features of Uogtag include: 1. **Optimizing graph topology**: Optimize the graph topology by selecting a graph learner suitable for the input data. 2. **Adaptive data augmentation**: Combine contrastive learning and adaptive data augmentation to improve the learning and applicability of the graph structure. 3. **Multi - view contrastive learning**: Maximize the mutual information between two different contrastive views (the learning view and the anchoring view) by constructing them, thereby optimizing the finally learned adjacency matrix \(\tilde{S}\). ### Specific methods 1. **Graph learner**: - **Graph learner I**: Applicable to input data without provided topological information, using full - graph parameterization (FGP) and two metric - learning - based graph learners (attention mechanism and multi - layer perceptron). - **Graph learner II**: Applicable to graph data with known initial topological information, using a graph - neural - network - based (GNNs) graph learner. 2. **Post - processing**: - **Sparsification**: Sparsify the preliminarily learned adjacency matrix \(S'\) by the k - nearest neighbor (KNN) method. - **Symmetrization and activation**: Ensure that the adjacency matrix is symmetric and non - negative. - **Normalization**: Normalize the edge weights of the adjacency matrix to the range of [0, 1] through symmetric normalization. 3. **Multi - view contrastive learning**: - **Graph view construction**: Define the learning view \(G_l = (\tilde{S}, X)\) and the anchoring view \(G_a=(A_a, X)\). - **Data augmentation**: Generate contrastive views through an adaptive data augmentation method, ignoring potential noisy edges. - **Contrastive learning**: Use node - level contrastive learning to maximize the mutual information between the two views. ### Experimental verification The paper verifies the robustness and effectiveness of Uogtag in dealing with noisy graphs and label - scarce situations through experiments on multiple real - world datasets. The experimental results show that Uogtag performs well in various graph structure learning tasks, especially under adversarial conditions and different parameter settings. In conclusion, by proposing the Uogtag framework, this paper effectively solves the main problems existing in the existing graph structure learning methods and improves the practicality and generalization ability of graph neural networks in various tasks.

Unsupervised Graph Structure Learning Based on Optimal Graph Topology Modeling and Adaptive Data Augmentation

Learning to Augment Graph Structure for Both Homophily and Heterophily Graphs

NGAT: Attention in Breadth and Depth Exploration for Semi-Supervised Graph Representation Learning

Towards Unsupervised Deep Graph Structure Learning

Efficient Topology-aware Data Augmentation for High-Degree Graph Neural Networks

GraphGLOW: Universal and Generalizable Structure Learning for Graph Neural Networks

A Survey on Graph Structure Learning: Progress and Opportunities

Harnessing Collective Structure Knowledge in Data Augmentation for Graph Neural Networks

Position-aware Structure Learning for Graph Topology-imbalance by Relieving Under-reaching and Over-squashing

A Unified Structure Learning Framework for Graph Attention Networks

Structural Imbalance Aware Graph Augmentation Learning

GaGSL: Global-augmented Graph Structure Learning via Graph Information Bottleneck

UGSL: A Unified Framework for Benchmarking Graph Structure Learning

Semi-Supervised Graph Structure Learning via Dual Reinforcement of Label and Prior Structure

OpenGraph: Towards Open Graph Foundation Models

Unsupervised Structure-Adaptive Graph Contrastive Learning.

Exploring Structure-Adaptive Graph Learning for Robust Semi-Supervised Classification

UniGraph: Learning a Unified Cross-Domain Foundation Model for Text-Attributed Graphs

Data Augmentation for Deep Graph Learning: A Survey

AAGCN: a graph convolutional neural network with adaptive feature and topology learning

Graph Data Augmentation for Node Classification