Ever logged into Facebook to see a friend recommendation, and vaguely recognise the name – it's your old friend from school! You haven't talked to them in ages, but here they are, popping up on your feed. How is it possible?
This is thanks to a powerful technology called Graph Neural Networks (GNNs). Before we start driving deep into graph neural networks, let us understand some basics of AI and how it boosts Social Network Analysis.
In Social Network Analysis, understanding neural networks is fundamental. The neural network is a subset of Artificial Intelligence (AI) inspired by the structure and operation of the human brain.
Imagine various nodes that are interconnected with each other to transmit and process information, identical to how neurons function in the human brain.
These nodes are the core elements of a neural network. Each node receives information from its neighbours, performs a simple function, transmits a signal to other connected nodes, and simultaneously keeps learning hidden patterns from data.
fig. 1. Comparing the human brain with artificial neural networks
Source: https://bit.ly/40fI622
To build an efficient neural network, it has to be exposed to a large amount of data. Machine Learning with Neural Networks enables us to train them on vast data and adjust the connections between neurons.
We can classify different types of neural networks based on the input data and output. Each type of neural network has a specific task and is often combined to create suitable solutions. For example, Convolutional Neural Networks excel at image recognition and video analysis, while Recurrent Neural Networks are trained to process sequential data.
These help to enhance user experience and provide personalised suggestions on social media platforms. Now, the question arises of how neural networks are used in social network analysis.
Social network analysis is a powerful method that visualises and analyses relationships and connections between entities or individuals within a network. Social Network Analysis maps nodes as the user and the edge connecting nodes as the relationships.
Traditional social network analysis techniques relied on various metrics like degree centrality which checks the number of connections a user has, clustering coefficient, which helps to verify how densely connected a group of users is, etc.
But as more data becomes abundant traditional social network analysis tools encounter issues like information overload, suboptimal recommendation performance, struggle with complex network structures, etc. This is where Graph Neural Networks (GNNs) come into play.
Graph Neural Networks are designed to work with graph-structured data like social networks. Machine Learning and Deep Learning tools can solve simple data structures with regular grids like text or images. But to interpret complex graph data structures without a fixed form with variable size and unordered nodes, like social media networks, we need GNN.
Fig 2. Comparing simple data with graph data
GNNs use a unique message-passing technique to unravel this complex graph data. In message passing, when a node passes information, it gives its solution as output and exchanges information about the nearest neighbours. This allows each node to learn about complex and hidden patterns within the network.
Fig 3. A graph neural network visualisation of the characters in the Game of Thrones.
Source: https://bit.ly/3Abxfvn
Graph data structures like social media network analysis hugely benefited from this approach. Graph neural network applications in Social Media Analysis include:
GNN is transforming social media networking, leveraging the power of network connections and relationships. As we keep exploring, we expect to witness further advancements in the power of networking.