Abstract:Evolving temporal networks serve as the abstractions of many real-life dynamic systems, e.g., social network and e-commerce. The purpose of temporal network embedding is to map each node to a time-evolving low-dimension vector for downstream tasks, e.g., link prediction and node classification. The difficulty of temporal network embedding lies in how to utilize the topology and time information jointly to capture the evolution of a temporal network. In response to this challenge, we propose a temporal motif-preserving network embedding method with bicomponent neighbor aggregation, named TME-BNA. Considering that temporal motifs are essential to the understanding of topology laws and functional properties of a temporal network, TME-BNA constructs additional edge features based on temporal motifs to explicitly utilize complex topology with time information. In order to capture the topology dynamics of nodes, TME-BNA utilizes Graph Neural Networks (GNNs) to aggregate the historical and current neighbors respectively according to the timestamps of connected edges. Experiments are conducted on three public temporal network datasets, and the results show the effectiveness of TME-BNA.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to simultaneously utilize topological structure and temporal information in dynamic networks to capture the evolution patterns of the network for effective node embedding. Specifically, the paper focuses on how to preserve temporal motifs in temporal network embedding and proposes a Temporal Motif - Preserving Network Embedding method based on two - component Neighbor Aggregation (TME - BNA). This method aims to capture the topological dynamics of nodes by constructing additional edge features and using Graph Neural Networks (GNNs) to aggregate historical and current neighbors, thereby improving the understanding and prediction ability of temporal networks. ### Main Contributions 1. **Construction of Additional Edge Features Based on Temporal Motifs**: - Introduced a method for constructing additional edge features based on temporal motifs, which can explicitly combine complex topological structure and temporal information. - Temporal motifs refer to small sub - graph patterns that frequently appear in temporal networks and reflect the evolution laws of temporal networks. 2. **Two - Component Neighbor Aggregation Based on Multi - Head Attention**: - Proposed a two - component neighbor aggregation method based on the multi - head attention mechanism, which aggregates the historical and current neighbors of nodes respectively. - This method can capture the recent changes of multi - hop neighbors, thus better reflecting the dynamic characteristics of nodes. 3. **Experimental Verification**: - Conducted experiments on three publicly available temporal network datasets and compared with existing temporal network embedding methods, and the results show that TME - BNA has superior performance. ### Method Overview 1. **Node Memory Update**: - Use Gated Recurrent Units (GRUs) to update the node's memory according to the interaction sequence to model the dynamic characteristics of the node. 2. **Edge Feature Construction Based on Temporal Motifs**: - Construct additional edge features by counting the temporal motif instances of each edge, which can explicitly combine topological structure and temporal information. 3. **Two - Component Neighbor Aggregation**: - Use Graph Neural Networks (GNNs) based on the multi - head attention mechanism to aggregate the historical and current neighbors of nodes respectively to capture the topological dynamics of neighbors. 4. **Link Prediction**: - Predict the interaction probability between nodes by training the model and use the cross - entropy loss function for optimization. ### Experimental Setup and Results - The paper conducted experiments on three publicly available temporal network datasets to verify the effectiveness of TME - BNA. - The experimental results show that TME - BNA outperforms existing temporal network embedding methods in multiple tasks (such as link prediction and node classification). ### Conclusion TME - BNA effectively captures the complex topological and temporal information in temporal networks by combining temporal motifs and the multi - head attention mechanism, providing a new solution for temporal network embedding. Future work can further explore more temporal motifs and more efficient neighbor aggregation methods.

TME-BNA: Temporal Motif-Preserving Network Embedding with Bicomponent Neighbor Aggregation

Motif-Preserving Temporal Network Embedding

Motif-Preserving Dynamic Attributed Network Embedding

MERIT: Learning Multi-level Representations on Temporal Graphs.

Learning dynamic embeddings for temporal attributed networks

TigeCMN: On exploration of temporal interaction graph embedding via Coupled Memory Neural Networks

Dynamic Network Embeddings for Network Evolution Analysis.

Learning Temporal Interaction Graph Embedding Via Coupled Memory Networks.

Temporal Network Embedding with Micro- and Macro-dynamics

Exploring Temporal Information for Dynamic Network Embedding

Temporal Motif-Based Attentional Graph Convolutional Network for Dynamic Link Prediction

TempNodeEmb:Temporal Node Embedding considering temporal edge influence matrix

TempME: Towards the Explainability of Temporal Graph Neural Networks via Motif Discovery

Multi-Aspect Temporal Network Embedding: A Mixture of Hawkes Process View

TE-DyGE: Temporal Evolution-Enhanced Dynamic Graph Embedding Network.

BP-MoE

Temporal Network Embedding Enhanced With Long-Range Dynamics and Self-Supervised Learning

Temporal Network Embedding for Link Prediction via VAE Joint Attention Mechanism

Exploring Evolution of Dynamic Networks Via Diachronic Node Embeddings

Embedding Temporal Network Via Neighborhood Formation