Graph Neural Networks: Link Prediction (Part II)

When It Comes to Forecasting Connections Within a Network

As many real-world problems can naturally be modeled as a network of nodes and edges, Graphical Neural Networks (GNNs) provide a powerful approach to solve them. By leveraging this inherent structure, they can learn more efficiently and solve complex problems where standard machine learning algorithms fail.

In this context, the first article of this series on GNNs discusses the benefits of such models when it comes to classifying entities in a network or predicting a continuous attribute.

But what if the task to be solved is about predicting the existence of a relationship between entities or a characteristic of this relationship?

Real-life examples are abundant:

• Retailers are interested in predicting the satisfaction score that consumers would give to their products in order to improve their recommendation tool.
• Social networks would find it useful to predict the likelihood that two users connect to improve their suggestions and ultimately help each of them expand your network.
• Chemists study the existence of interactions between molecules in order to discover new drugs or to avoid unexpected side effects when…