How GNNs can help in finding hidden risks in Supply Chains?
Being able to predict interconnections between suppliers can help in detecting anomalies for financial fraud detection and various other types of hidden risks that may not be obvious from financial and public stakeholder data.
Machine Learning is a powerful tool for dealing with complex patterns. However, traditional techniques are limited in analyzing interactions between entities and the network structure, which is very important for calculating systemic risk [1]. Over the past decade, researchers have developed neural networks that operate on graph data called graph neural networks, or GNNs. Recent developments have increased their capabilities and expressive power. Practical applications in areas such as antibacterial discovery, physics simulations, fake news detection, traffic prediction, and recommendation systems [2] are being tested.
In this study, GNNs were tested to analyze systemic risk. GNNs were tested against traditional ML algorithms in classifying entities by systemic risk importance in two different networks, comparing their generalization capabilities with different amounts of available data. GNNs achieve a 94% and 15% Matthew’s Correlation Coefficient (MCC) average percentage increase (a measure of the quality of binary classifications) compared to ML, achieving statistically significant MCC improvements in most scenarios.
In this article, I explore how GNNs work and how they can be used to find hidden risks in global supply chains.
This is a part of a series of articles I am writing about powerful technologies in Artificial Intelligence that can minimize supplier risks in the supply chain. For the past few months, I have talked to many professionals in the Oil and Gas Supply Chain to design an AI-powered supplier risk platform that can increase visibility into hidden risks as well as broaden supplier risk management to save up to 15% of costs [source] and reduce emissions by 20% [source].
How does a GNN work?
A Graph Neural Network can analyze and find patterns in graph data. Before going further, we should define what I mean by a graph.
Graphs
A graph is a structure representing entities and their relations.
It consists of two sets: nodes and edges, with nodes being fixed entities with properties (embeddings) and the edges representing the relations between the nodes. These edges can be directed (from one node to another) capturing asymmetric relations between nodes or undirected, capturing symmetric relations. As a graph maps relationships between various entities, they can be incredibly useful in finding hidden relationships that can pose risks in supplier data.
Graphs can be weighted or unweighted. In a weighted graph, each edge is associated with a scalar weight. For example, such weights might represent lengths or connectivity strengths.
Graphs can also be either homogeneous or heterogeneous. In a homogeneous graph, all the nodes represent instances of the same type and all the edges represent relations of the same type. For instance, a social network is a graph consisting of people and their connections, representing the same entity type.
In contrast, in a heterogeneous graph, the nodes and edges can be of different types. For instance, the graph encoding a marketplace will have buyer, seller, and product nodes that are connected via wants-to-buy, has-bought, is-customer-of, and is-selling edges. [3]
Here are some examples of real-world objects that can be represented as graphs: Molecules (with atoms as nodes and their bonds as edges) and citation networks (each paper is a node, and each directed edge is a citation between one paper and another).
What types of problems can GNNs solve?
There are three types of prediction tasks we can perform on GNNs:
- Graph-level: Predicting a property for the whole graph. Examples: Image classification, Sentiment Analysis, etc.
- Node-level: Predicting the identity or role of each node. In the context of supply chains, this can be used to predict the risk levels of individual suppliers.
- Edge-level: Predicting the relationships between nodes. This can be used to predict which suppliers are at a chance of merging, for example. This prediction can be diversified based on the dataset.
Through NLP, we can use external data sources such as social media, customer feedback, stakeholder reports, etc. as embeddings for nodes, along with internal financial and logistical data sources, to find patterns and connections among a breadth of parameters and risks.
Stay tuned for a tutorial on applying GNNs to visualize supplier risk!
Relevant articles
Sources:
[1]
[2]
[3]
