Node embedding technique

As explained in Graph Algorithms for Data Science

A node-embedding algorithm computes vector representations of graph nodes that encode structural or relational properties.

A node embedding technique learns a compact vector for each graph node instead of feeding the entire adjacency matrix into a model. The vectors encode structural or relational properties, so nearby points in embedding space represent nodes with similar neighborhood topology. section 9.0

The embedding step is unsupervised: it learns representations without labeled training data. Models use deep learning and nonlinear dimensionality reduction, while the embedding dimension controls the number of elements in each vector. section 9.0

Node embeddings address practical limitations of adjacency-matrix features:

• Reduce the number of input features and limit overfitting risk. section 9.0
• Keep the representation size stable when nodes are added or removed. section 9.0
• Replace painstaking manual feature engineering with a learned representation. section 9.0

Figure 9.2 Adjacency matrix

Use the adjacency matrix illustration to see the high-dimensional input that node embeddings compress. section 9.0

How Graph Algorithms for Data Science covers it

How deep it goes

Defines the learning objective and dimensionality reduction, but stops short of explaining a specific embedding algorithm or implementation.

Examples

Contrasts adjacency-matrix features with embeddings using a graph containing one million nodes.

What it assumes

Assumes graph and machine-learning framing while explaining adjacency-matrix limitations from the representation problem.

What you can do after

Evaluate/understand why learned low-dimensional node vectors can replace manually engineered graph features.

How it connects

Connects adjacency matrices, dimensionality reduction, machine learning, deep learning, and feature engineering.

liveBookmanning.com

Read Graph Algorithms for Data Science on liveBook Get the book at manning.com

This explainer was generated by Manning's knowledge graph from the text of Graph Algorithms for Data Science. The section links point to the parts of the book it draws on.