Link Prediction Lab: Node Embeddings & Edge Scoring

What Link Prediction Is

Link prediction is a fundamental problem in network analysis where the goal is to predict missing or future links between nodes. This task is crucial for applications like social network analysis, recommendation systems, and biological networks.

In this simulation, you can observe how node embeddings are generated using graph convolutional networks (GCNs) and how these embeddings are used to score potential edges.

How Node Embeddings Work

Node embeddings represent each node in a network as a vector of real numbers. These vectors capture the structural and attribute information about nodes, allowing them to be compared and grouped based on their similarity.

The neighbor-mean aggregation method used here averages the embeddings of a node's neighbors to update its own embedding, iteratively refining these representations.

Edge Scoring Methods

To score potential edges between nodes, this simulation uses two common methods: cosine similarity and dot product. Both methods measure how similar the embeddings of two nodes are.

Cosine similarity measures the cosine of the angle between two vectors, while the dot product directly multiplies corresponding components of the vectors.

Evaluating Predictions

The performance of link predictions is typically evaluated using metrics like Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC). In this simulation, AUC tracks how well predicted links match actual held-out edges.

By adjusting parameters such as aggregation mix rate and score threshold, you can observe changes in AUC and understand their impact on prediction accuracy.

Frequently asked questions

Try it live

Everything above runs in your browser — open Link Prediction Lab: Node Embeddings & Edge Scoring and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.