However, the availability of such resources remains scarce and evaluation faces added complexity due to reproducibility issues in experimental protocols.

Node centralities play a pivotal role in network science, social network analysis, and recommender systems. In temporal data, static path-based centralities like closeness or betweenness can give misleading results about the true importance of nodes in a temporal graph. To address this issue, temporal generalizations of betweenness and closeness have been defined that are based on the shortest time-respecting paths between pairs of nodes.

We present the T emporal Graph Benchmark (TGB), a collection of challenging and diverse benchmark datasets for realistic, reproducible, and robust evaluation of machine learning models on temporal graphs.

WhenT is larger, more recent edges are assignedsmallDAmagnitudes,sothattheessentialsemantic information is preserved. This theorem guarantees that our DA techiniques do not break the original edge time distribution. There are 4,066 drop-out events (= 0.98%). Based on the validation results, using two TGAT layers and two attention heads with dropout rate of 0.1 gives the best performance. For inference, we inductively compute the embeddings for both the unseen and observed nodes at each time point that the graph evolves, or when the node labels are updated.