A graph neural network is a deep learning method that analyzes and makes predictions based on data described by a graph. Deep learning is a machine learning (ML) technique that teaches computers to mimic the workings of a human brain. A graph in computer science, meanwhile, is a data structure that has two parts—vertices and edges. We’ll discuss these components in greater detail in another section.

Graphs in graph neural networks, however, differ from those we see and learn about in math classes. They aren’t limited to pie charts, bar graphs, and others like them. You’ll see why later on.

Other interesting terms…

We mentioned the graph components analyzed by graph neural networks above. Let’s go into them now.

### What Components Do Graph Neural Networks Analyze?

Graph neural networks analyze graphs’ vertices and edges.

Vertices or nodes are the points where two data points intersect or share a commonality in a dataset undergoing analysis. Edges, on the other hand, are unshared points by two data points. The image below illustrates the two graph parts.

### What Can Graph Neural Networks Analyze?

As mentioned earlier, graph neural networks can analyze more than the various charts we can create in spreadsheet applications like Microsoft Excel. They can analyze images, text, social networks, citations, and more.

We’ll go into how graph neural networks process each data type below.

#### Images

In such a case, a machine translates each image pixel into vertices and edges in a graph. Each component can represent the RGB value of the pixel. Given a set of images, graph neural networks can see which color is most often used by specific artists.

#### Text

In the text, each letter can be considered either a vertex or an edge, depending on where it appears in each word. Graph neural networks can then analyze how many times the letter “a,” for instance, is used at the beginning of words given a vast number of books.

#### Social Network

Graph neural networks can also scrutinize the relationships between a given set of people on social media. People with various connections, in that case, would be vertices, while those with only one connection to someone else are edges.

#### Citation

If you want to know how many times the class members used a specific set of reference books, you can also use a graph neural network, with books cited by more than one student appearing as vertices and those used by a single student as edges.

### What Are Graph Neural Networks Used For?

A graph neural network can be helpful in computer vision, natural language processing (NLP), traffic analysis, and chemistry.

#### Computer Vision

Graph neural networks can help computers distinguish and identify objects in images and videos. While today’s machines have yet to see images and videos the way humans do, more advanced graph neural networks can solve some image classification problems.

#### NLP

In the realm of NLP, where computers analyze human language to enable human-to-computer communication and vice versa, graph neural networks enable text classification, exploiting semantics in machine translation, user geolocation, relation extraction, and question answering.

#### Traffic Analysis

Graphic neural networks also help forecast traffic speed and volume or density for smart transportation systems.

#### Chemistry

Chemists use graph neural networks to analyze the structure of molecules or compounds. The graphs they use comprise two parts–the atoms are the vertices while the chemical bonds are the edges.

To date, computer scientists have tried applying graph neural networks for program verification, program reasoning, social influence prediction, recommender systems, electronic health record (EHR) modeling, brain networks, and adversarial attack prevention.

Uber Eats uses the GraphSage network to recommend restaurants and food to users. Alibaba, meanwhile, uses self-developed Aligraph to promote products. Pinterest and Amazon also have their own graph neural networks for user recommendations. The Google Brain team employs graph neural networks to optimize hardware chips.

Another example is SuperGlue, which aids in three-dimensional (3D) image reconstruction, place recognition, localization, and mapping. There are tons more.

Graph neural networks can help solve any problem whose components can be represented by graphs. As you’ve seen, the tech already powers many real-world applications, but more will benefit from it in the future.