What are Graph Embeddings?
Learn about the numerical vector representation of graphs.
Embeddings
As we've seen, graphs collect and store complex, real-life interactions. This makes graphs robust data structures that are intuitive and flexible. Since graphs are non-euclidean data structures, they can't be used directly as input in a machine learning algorithm. This is why we need to learn graph embeddings in a low-dimensional space. We can perform different types of graph analytics tasks using embeddings. Embeddings are vector representations of a graph in an
Graph embeddings are usually a vector with float values, whereas the user predetermines the size. In the example above, every node is represented by an embedding size of 7
. For instance, if we have a graph with 20 nodes, the resulting embedding matrix shape would be (20,7)
. The numbers capture the geometric relationships in the original graph.
Representation learning
The process of learning graph embeddings is also called graph representation learning. Formally, this can be defined as a task to learn a mapping function , where ...