Abstract:Graph representation learning, aiming to learn low-dimensional representations which capture the geometric dependencies between nodes in the original graph, has gained increasing popularity in a variety of graph analysis tasks, including node classification and link prediction. Existing representation learning methods based on graph neural networks and their variants rely on the aggregation of neighborhood information, which makes it sensitive to noises in the graph. In this paper, we propose Graph Denoising Policy Network (short for GDPNet) to learn robust representations from noisy graph data through reinforcement learning. GDPNet first selects signal neighborhoods for each node, and then aggregates the information from the selected neighborhoods to learn node representations for the down-stream tasks. Specifically, in the signal neighborhood selection phase, GDPNet optimizes the neighborhood for each target node by formulating the process of removing noisy neighborhoods as a Markov decision process and learning a policy with task-specific rewards received from the representation learning phase. In the representation learning phase, GDPNet aggregates features from signal neighbors to generate node representations for down-stream tasks, and provides task-specific rewards to the signal neighbor selection phase. These two phases are jointly trained to select optimal sets of neighbors for target nodes with maximum cumulative task-specific rewards, and to learn robust representations for nodes. Experimental results on node classification task demonstrate the effectiveness of GDNet, outperforming the state-of-the-art graph representation learning methods on several well-studied datasets. Additionally, GDPNet is mathematically equivalent to solving the submodular maximizing problem, which theoretically guarantees the best approximation to the optimal solution with GDPNet.

Long-short Distance Aggregation Networks for Positive Unlabeled Graph Learning

Learning Graph Neural Networks with Positive and Unlabeled Nodes

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

Positive-Unlabeled Node Classification with Structure-aware Graph Learning

Towards Robust Graph Neural Networks against Label Noise

Positive-Unlabeled Learning for Network Link Prediction

Unraveling the Impact of Heterophilic Structures on Graph Positive-Unlabeled Learning

GAIN: Graph Attention & Interaction Network for Inductive Semi-Supervised Learning over Large-scale Graphs

ANGraph: Attribute-Interactive Neighborhood-Aggregative Graph Representation Learning

Multi-graph aggregated graph neural network for heterogeneous graph representation learning

NP$^2$L: Negative Pseudo Partial Labels Extraction for Graph Neural Networks

AGNet: An Attention-Based Graph Network for Point Cloud Classification and Segmentation

LSPI: Heterogeneous Graph Neural Network Classification Aggregation Algorithm Based on Size Neighbor Path Identification

GraphAIR: Graph representation learning with neighborhood aggregation and interaction

CAGNN: Cluster-Aware Graph Neural Networks for Unsupervised Graph Representation Learning

Multi-Relational Graph Representation Learning with Bayesian Gaussian Process Network.

DEER: Distribution Divergence-based Graph Contrast for Partial Label Learning on Graphs

MHNF: Multi-hop Heterogeneous Neighborhood information Fusion graph representation learning

HGAMLP: Heterogeneous Graph Attention MLP with De-Redundancy Mechanism

Learning Robust Representations with Graph Denoising Policy Network

Deep pairwise learning for user preferences via dual graph attention model in location-based social networks