FILDNE: A Framework for Incremental Learning of Dynamic Networks Embeddings

Representation learning on graphs has emerged as a powerful mechanism to automate feature vector generation for downstream machine tasks. The advances in representation have centered both homogeneous and heterogeneous graphs, where the latter presenting challenges associated with multi-typed nodes and/or edges. In this paper, we consider additional challenge of evolving graphs. We ask question whether static can be leveraged dynamic how? It is important able incorporate those maximize utility generalization methods. To that end, propose Framework Incremental Learning Dynamic Networks Embedding (FILDNE), which utilize any existing method node embeddings while keeping computational costs low. FILDNE integrates vectors computed using standard methods over different timesteps into single by developing convex combination function alignment mechanism. Experimental results several tasks, seven real-world datasets, show reduce memory (up 6x) time 50x) providing competitive quality measure gains (e.g., improvements up 19 pp AUC link prediction 33 mAP graph reconstruction) respect contemporary