Streaming Graph Embeddings via Incremental Neighborhood Sketching

Graph embeddings have become a key paradigm to learn node representations and facilitate downstream graph analysis tasks. Many real-world scenarios such as online social networks communication involve streaming graphs, where edges connecting nodes are continuously received in manner, making the underlying structures evolve over time. Such raises great challenges for embedding techniques not only capturing structural dynamics of graph, but also efficiently accommodating high-speed edge streams. Against this background, we propose SGSketch, highly-efficient technique via incremental neighborhood sketching. SGSketch cannot generate high-quality from by gradually forgetting outdated edges, update generated an updating mechanism. Our extensive evaluation compares against sizable collection state-of-the-art using both synthetic graphs. The results show that achieves superior performance on different tasks, showing 31.9% 21.9% improvement average best-performing static dynamic baselines, respectively. Moreover, is significantly more efficient learning processes, 54x-1813x 118x-1955x speedup baseline techniques,