Abstract:Temporal graph representation learning aims to generate low-dimensional dynamic node embeddings to capture temporal information as well as structural and property information. Current representation learning methods for temporal networks often focus on capturing fine-grained information, which may lead to the model capturing random noise instead of essential semantic information. While graph contrastive learning has shown promise in dealing with noise, it only applies to static graphs or snapshots and may not be suitable for handling time-dependent noise. To alleviate the above challenge, we propose a novel Temporal Graph representation learning with Adaptive augmentation Contrastive (TGAC) model. The adaptive augmentation on the temporal graph is made by combining prior knowledge with temporal information, and the contrastive objective function is constructed by defining the augmented inter-view contrast and intra-view contrast. To complement TGAC, we propose three adaptive augmentation strategies that modify topological features to reduce noise from the network. Our extensive experiments on various real networks demonstrate that the proposed model outperforms other temporal graph representation learning methods.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is: **How to reduce the influence of temporal - dependence noise in temporal graph representation learning, thereby improving the accuracy of future interaction prediction**. Specifically, existing temporal graph representation learning methods usually focus on capturing fine - grained information, which may cause the model to capture random noise rather than essential semantic information. Moreover, although contrastive learning (CL) has shown potential in dealing with noise in static graphs or snapshots, it may not be suitable for dealing with temporal - dependence noise. To solve these problems, the authors propose a new model named **Temporal Graph representation learning with Adaptive augmentation Contrastive (TGAC)**. This model performs adaptive augmentation by combining prior knowledge and temporal information and constructs the model by defining the contrastive objective function between augmented views. The following are the main contributions of this model: 1. **Introduction of temporal information**: Temporal information is introduced in the edge generation process, enabling the model to better capture the evolution characteristics of the graph structure, thereby improving representation learning. 2. **Proposal of a new temporal graph augmentation method**: The temporal and structural information of nodes and their neighborhoods is used for augmentation, which not only preserves important temporal features but also enhances the original graph. 3. **Design of a graph pruning scheme**: Edge - centrality measures are used to remove noisy or redundant connections to enhance important topological structures and improve node representations. 4. **Experimental verification**: The experimental results show that TGAC outperforms other state - of - the - art temporal graph representation learning models in tasks such as link prediction and node classification. ### Formula Summary - **Edge - centrality calculation formula**: - For undirected graphs: \[ \phi_t^{uv}=\frac{\varphi(u)+\varphi(v)}{2}+\alpha t_{uv} \] - For directed graphs: \[ \phi_t^{uv}=\varphi(v)+\alpha t_{uv} \] - **Edge - removal probability formula**: \[ p_t^{uv}=\min\left(\frac{w_t^{\max}-w_t^{uv}}{w_t^{\max}-\mu_t^w}\cdot p_e, p_r\right) \] where \( w_t^{uv}=\log(\phi_t^{uv}) \), \( p_e \) is a hyperparameter that controls the overall removal probability, and \( p_r < 1 \) is a cutoff probability to prevent the removal probability from being too high. - **Total loss function**: \[ L = \lambda L_{\text{task}}+L_{\text{cl}} \] where \( L_{\text{task}} \) is the task loss function, \( L_{\text{cl}} \) is the contrastive loss function, and \( \lambda \) is a hyperparameter that balances the weights of the two loss functions. These formulas and methods work together to enable the TGAC model to effectively reduce the noise influence in temporal graph representation learning and improve the robustness and generalization ability of the model.

Temporal Graph Representation Learning with Adaptive Augmentation Contrastive

Time-aware Graph Structure Learning Via Sequence Prediction on Temporal Graphs.

Full-Attention Driven Graph Contrastive Learning: with Effective Mutual Information Insight

Graph Contrastive Learning with Adaptive Augmentation

Spatial-Temporal Graph Learning with Adversarial Contrastive Adaptation

Spatio-Temporal Meta Contrastive Learning

Triplet teaching graph contrastive networks with self-evolving adaptive augmentation

Graph Contrastive Learning with Cohesive Subgraph Awareness

Learning dynamic graph representations through timespan view contrasts

Features Based Adaptive Augmentation for Graph Contrastive Learning

Graph Contrastive Learning with Augmentations

LAC: Graph Contrastive Learning with Learnable Augmentation in Continuous Space

Contrastive Graph Convolutional Networks with adaptive augmentation for text classification

Contrastive representation learning on dynamic networks

Graph Contrastive Learning with Constrained Graph Data Augmentation

Graph Contrastive Learning with Generative Adversarial Network

Self-Supervised Dynamic Graph Representation Learning via Temporal Subgraph Contrast

Spectral Augmentations for Graph Contrastive Learning

Augmentation-Free Graph Contrastive Learning of Invariant-Discriminative Representations

Graph Contrastive Learning with Reinforced Augmentation

Graph Representation Learning via Contrasting Cluster Assignments