CoANE: Modeling Context Co-occurrence for Attributed Network Embedding

Attributed network embedding (ANE) is to learn low-dimensional vectors so that not only the structure but also node attributes can be preserved in space. Existing ANE models do consider specific combination between graph and attributes. While each has its structural characteristics, such as highly-interconnected neighbors along with their certain patterns of attribute distribution, node's neighborhood should depicted by multi-hop nodes, clusters or social circles. To model information, this paper, we propose a novel model, Context Co-occurrence-aware Network Embedding (CoANE). The basic idea CoANE context involved diverse patterns, apply convolutional mechanism encode positional information treating channel. learning co-occurrence capture latent circles node. better semantic knowledge devise three-way objective function, consisting positive likelihood, contextual negative sampling, reconstruction. We conduct experiments on five real datasets tasks link prediction, label classification, clustering. results exhibit significantly outperform state-of-the-art models.