2018/11/26 by Dimitris Berberidis, Georgios B. Giannakis, Berberidis, Dimitris +1
Computer Science · Physics and Astronomy · #Advanced Graph Neural Networks #Complex Network Analysis Techniques #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Opinion Dynamics and Social Influence
paper · pdf · doi:10.48550/arxiv.1811.10797
openalex publication_date 2018/11/26 · openalex created_date 2022/08/01 · openalex updated_date 2026/07/28
Node embedding is the task of extracting informative and descriptive features\nover the nodes of a graph. The importance of node embeddings for graph\nanalytics, as well as learning tasks such as node classification, link\nprediction and community detection, has led to increased interest on the\nproblem leading to a number of recent advances. Much like PCA in the feature\ndomain, node embedding is an inherently \unsupervised task; in lack of\nmetadata used for validation, practical methods may require standardization and\nlimiting the use of tunable hyperparameters. Finally, node embedding methods\nare faced with maintaining scalability in the face of large-scale real-world\ngraphs of ever-increasing sizes. In the present work, we propose an adaptive\nnode embedding framework that adjusts the embedding process to a given\nunderlying graph, in a fully unsupervised manner. To achieve this, we adopt the\nnotion of a tunable node similarity matrix that assigns weights on paths of\ndifferent length. The design of the multilength similarities ensures that the\nresulting embeddings also inherit interpretable spectral properties. The\nproposed model is carefully studied, interpreted, and numerically evaluated\nusing stochastic block models. Moreover, an algorithmic scheme is proposed for\ntraining the model parameters effieciently and in an unsupervised manner. We\nperform extensive node classification, link prediction, and clustering\nexperiments on many real world graphs from various domains, and compare with\nstate-of-the-art scalable and unsupervised node embedding alternatives. The\nproposed method enjoys superior performance in many cases, while also yielding\ninterpretable information on the underlying structure of the graph.\n