Deep node ranking for neuro?symbolic structural node embedding and classification

Network node embedding is an active research subfield of complex network analysis. This paper contributes a novel approach to learning embeddings and direct classification using ranking scheme coupled with autoencoder-based neural architecture. The main advantages the proposed Deep Node Ranking (DNR) algorithm are competitive or better performance, significantly higher speed lower space requirements when compared state-of-the-art approaches on 15 real-life benchmarks. Furthermore, it enables exploration relationship between symbolic derived sub-symbolic representations, offering insights into learned structure. To avoid complexity bottleneck in setting, DNR computes stationary distributions personalized random walks from given nodes mini-batches, scaling seamlessly larger networks. laws associated were also investigated 1488 synthetic Erd\H{o}s-R\'enyi networks, demonstrating its scalability tens millions links.