A Survey on Embedding Dynamic Graphs

Abstract

Embedding static graphs in low-dimensional vector spaces plays a key role in network analytics and inference, supporting applications like node classification, link prediction, and graph visualization.However, many real-world networks present dynamic behavior, including topological evolution, feature evolution, and diffusion. Therefore, several methods for embedding dynamic graphs have been proposed to learn network representations over time, facing novel challenges, such as time-domain modeling, temporal features to be captured, and the temporal granularity to be embedded. In this survey, we overview dynamic graph embedding, discussing its fundamentals and the recent advances developed so far. We introduce the formal definition of dynamic graph embedding, focusing on the problem setting and introducing a novel taxonomy for dynamic graph embedding input and output.We further explore different dynamic behaviors that may be encompassed by embeddings, classifying by topological evolution, feature evolution, and processes on networks. Afterward, we describe existing techniques and propose a taxonomy for dynamic graph embedding techniques based on algorithmic approaches, from matrix and tensor factorization to deep learning, random walks, and temporal point processes. We also elucidate main applications, including dynamic link prediction, anomaly detection, and diffusion prediction, and we further state some promising research directions in the area.

Sparse-Dyn: Sparse Dynamic Graph Multi-representation Learning via Event-based Sparse Temporal Attention Network

“…The former includes only structural information, while the latter includes another critical parameter: time. The dynamic continuous representation contains node interactions [26], and timestamped edges [12], where instantaneous events are recorded into the raw graph representations, such as creation and removal of nodes and edges.…”

“…For representation learning, the raw representations should be prepossessed to dynamic graphs first. Current researches mainly focus on two dynamic graphs: the continuous and discrete graphs [9,10,12,13]. The dynamic continuous graphs store the most information by projecting the raw representations to a 2D temporal graph, which is also a specific static graph appended with temporal information [11].…”

“…In order to deal with the complicated time-varied graphs, it is necessary and crucial to preprocess the raw dynamic graph representations, which record all continuous evolution of the graph over time, such as node emerging/disappearing and link addition/deletion [9,10,11,12,13]. Current researches [14,15,16,17,18,19,20] refine the raw dynamic representations to two main branches, dynamic continuous and discrete graphs.…”

“…In terms of the latter, discrete graphs, the structural representations are sampled to graph snapshots at regular time intervals, such as one day, over time. Although the developing networks are easier than the continuous ones, the temporal information is lost much more [12]. More details about time encoding approaches are discussed in Appendix 7.1.…”

Recommender systems based on graph embedding techniques: A comprehensive review

“…Graph embedding [31,[130][131][132] bridges the gap of the utilization of non-Euclidean data in machine learning methods for downstream tasks, including node classification [33,34], graph classification [35][36][37][38][39], link prediction [40][41][42][43], clustering [44] and stuff. Specifically, as for building the feature vectors of data instances as inputs required by machine learning methods [133], typical representation learning techniques, such as the artificially given methods [134] implementing based on hand-engineering with expert knowledge and the bag-of-words methods [135], are usually constrained to Euclidean data.…”