Graph Neural Network
- Introduced by
- Kipf, T.N. & Welling, M.
- Year
- 2017
- Type
- Deep learning on graph-structured data
Scientific review pending.
Sections asserting universal GNN sample-size thresholds are withheld pending scientific review. The original study does not establish a general eligibility threshold. Comparison sourceWhat does it do?
A Graph Neural Network (GNN) is a deep learning method, popularised by Kipf and Welling in 2017 with the Graph Convolutional Network, that learns from the relationships in network (graph) structures made of nodes and edges.
It is designed for data that is naturally relational, such as social networks, molecular structures, and recommendation systems.
The intuition behind the method
Many kinds of data are not flat tables of independent rows but webs of connected entities: people linked by friendships, atoms linked by bonds, items linked by who buys them together. A GNN takes that connectivity seriously. Instead of looking at each node in isolation, it lets every node listen to its neighbours and update its own representation based on what surrounds it, so the structure of the graph itself becomes part of what the model learns.
History
Graph Neural Networks moved into the mainstream with Kipf and Welling's 2017 Graph Convolutional Network, which framed learning on graphs as a form of neighbourhood-based convolution. Veličković and colleagues extended the idea in 2018 with Graph Attention Networks, letting nodes weigh their neighbours differently. Hamilton's 2020 monograph Graph Representation Learning consolidated the field, and GNNs have since become a standard tool for relational data.
When to use it
This section awaits scientific review. Do not use its conditions as method selection rules.
Original text awaiting review
Limitations and considerations
- The source does not specify limitations; suitability requires a separate assessment.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Learns directly from relational structure rather than forcing data into flat rows.
- Naturally suited to social networks, molecular structures, and recommendation systems.
- Assumption-light on feature distributions; handles continuous, categorical, and binary node features.
- Captures how a node's context — its neighbourhood in the graph — shapes its behaviour.
Applications
- Node and link analysis in social networks.
- Property prediction over molecular structures.
- Recommendation systems built on user–item interaction graphs.
- Explaining and learning relationships in any naturally connected dataset.
How it works
How it works
Formula
h_v^{(l+1)} = \sigma\!\left(\sum_{u \in \mathcal{N}(v)} \frac{1}{c_{vu}} W^{(l)} h_u^{(l)}\right)Related methods
Sources
Kipf, T.N. & Welling, M. (2017). Semi-Supervised Classification with Graph Convolutional Networks. ICLR.
Open publication referenceCopied.Veličković, P. et al. (2018). Graph Attention Networks. ICLR.
Open publication referenceCopied.Hamilton, W.L. (2020). Graph Representation Learning. Morgan & Claypool.
Open DOI referenceCopied.