Multi-Scale attributed node embedding

Abstract We present network embedding algorithms that capture information about a node from the local distribution over attributes around it, as observed random walks following an approach similar to Skip-gram. Observations neighbourhoods of different sizes are either pooled (AE) or encoded distinctly in multi-scale (MUSAE). Capturing attribute-neighbourhood relationships multiple scales is useful for range applications, including latent feature identification across disconnected networks with features. prove theoretically matrices node-feature pointwise mutual implicitly factorized by embeddings. Experiments show our computationally efficient and outperform comparable models on social web graphs.