GraphSage is a framework for inductive representation learning on large graphs, and is used to generate low-dimensional vector representations for nodes.
Some of the motivation behind GraphSage includes the idea that low-dimensional vector embeddings of nodes in large graphs have applications in machine learning. GraphSage is an inductive framework that leverages node attribute information to generate representations on previously unseen data. The framework trains on examples of graphs and a set of graphs, and afterwards can be used to generate node embeddings for previously unseen nodes or new input graphs.
The code that the framework utilizes is implemented in TensorFlow and can be easily integrated into other machine learning pipelines. GraphSage also has support for training on smaller, static graphs and graphs that don't have node features. The algorithm and paper are focused on the task of inductive generalization. It offers optional identity features that can be used with or without other node attributes.
GraphStage is intended for use on large graphs with more than 100,000 nodes. The overhead of sub-sampling then begins to outweigh the benefits of smaller graphs.
