Multi-relational graph convolutional network (GCN) in risk prediction

Patent metadata

Field	Value
Publication number	US-12437193-B2
Application number	US-202117241790-A
Country	US
Kind code	B2
Filing date	Apr 27, 2021
Priority date	Apr 27, 2021
Publication date	Oct 7, 2025
Grant date	Oct 7, 2025

Abstract

A graph neural network can be built and trained to predict a risk of an entity. A multi-relational graph network can include a first graph network and a second graph network. The first graph network can include a first set of nodes and a first set of edges connecting some of the nodes in the first set. The second graph network can include a second set of nodes and a second set of edges connecting some of the nodes in the second set. The first set of nodes and the second set of nodes can represent entities, the first set of edges can represent a first relationship between the entities and the second set of edges can represent a second relationship between the entities. A graph convolutional network (GCN) can be structured to incorporate the multi-relational graph network, and trained to predict a risk associated with a given entity.

First claim

What is claimed is:

1. A computer-implemented method of building a graph neural network to predict a risk on entities, comprising:
creating a multi-relational graph network including at least a first graph network including a first set of nodes and a first set of edges connecting at least some nodes in the first set of nodes, a second graph network including a second set of nodes and a second set of edges connecting at least some nodes in the second set of nodes, the first set of nodes and the second set of nodes representing entities, the first set of edges representing a first relationship between the entities and the second set of edges representing a second relationship between the entities;
structuring a graph convolutional network (GCN) that incorporates the multi-relational graph network;
training the GCN to predict a risk associated with a given entity,
wherein the trained GCN predicts the risk on the given entity due to climate,
wherein for each node in the first set of nodes and the second set of nodes, a node feature for the GCN includes a climate change impact indicator representing an impact of a climate change over a period on an entity represented by the node, the first set of edges represents online searches performed between the entities, and the second set of edges represents a correlation of equity returns between the entities,
wherein an edge connecting one entity and another entity in the first set of edges, represents a number of instances of directional searches performed on said one entity and said another entity immediately within a specified time period,
wherein the training including capturing interaction of climate variables and equity prices and transmission of climate risk to equity returns, the climate variables including variables corresponding to acute climate hazards and chronic climate hazards, and inputting the captured interaction of the climate variables and the equity prices and transmission of the climate risk to equity returns as values of the GCN's node feature corresponding to the climate change impact indicator;
receiving, via a user interface, a selection of the given entity whose risk due to climate is to be predicted;
in response to the receiving of the selection, running the trained GCN to predict the risk for the given entity based on the climate risk of the entities in the trained GCN; and
displaying the predicted risk for the given entity on the user interface.

2. The method of claim 1 , wherein said structuring the graph convolutional network (GCN) that incorporates the multi-relational graph network includes representing the first set of edges in a first adjacency matrix, representing the second set of edges in a second adjacency matrix, and aggregating the first adjacency matrix and the second adjacency matrix and using the aggregated adjacency matrices in computing node embeddings in the graph convolutional network.

3. The method of claim 2 , further including infusing sparsity into at least one of the first adjacency matrix and the second adjacency matrix.

4. The method of claim 3 , wherein infusing sparsity includes removing some of the edges in at least one of the first set of edges and the second set of edges based on a predefined threshold.

5. The method of claim 3 , wherein infusing sparsity includes randomly dropping some edges in at least one of the first set of edges and the second set of edges at each training epoch of the GCN.

6. The method of claim 1 , wherein said structuring the graph convolutional network (GCN) that incorporates the multi-relational graph network includes forming a supra-graph including adjacency matrix A and using the adjacency matrix A in computing node embeddings in the graph convolutional network, wherein

A
=

[

A

(
1
)

I

I

A

(
2
)

]

,

wherein A (1) represents the first set of edges, A (2) represents the second set of edges, and I represents an identity matrix.

7. The method of claim 6 , further including infusing sparsity into at least one of A (1) and A (2) .

8. The method of claim 7 , wherein infusing sparsity includes removing some edges represented in A (2) and A (1) based on a predefined threshold.

9. The method of claim 7 , wherein infusing sparsity includes randomly dropping some edges represented in A (2) and A (1) at each training epoch of the GCN.

10. The method of claim 1 , wherein the GCN predicts a credit risk on an entity.

11. The method of claim 10 , wherein for each node in the first set of nodes and the second set of nodes, a node feature for GCN includes a score representing equity ratios of an entity represented by the node, the first set of edges represents online searches performed between the entities, and the second set of edges represents a correlation of equity returns between the entities.

12. A system for building a graph neural network to predict a risk on entities, comprising:
a processor; and
a memory device coupled with the processor;
the processor configured to:
create a multi-relational graph network including at least a first graph network including a first set of nodes and a first set of edges connecting at least some nodes in the first set of nodes, a second graph network including a second set of nodes and a second set of edges connecting at least some nodes in the second set of nodes, the first set of nodes and the second set of nodes representing entities, the first set of edges representing a first relationship between the entities and the second set of edges representing a second relationship between the entities;
structure a graph convolutional network (GCN) that incorporates the multi-relational graph network;
train the GCN to predict a risk associated with a given entity,
wherein the trained GCN predicts the risk on the given entity due to climate,
wherein for each node in the first set of nodes and the second set of nodes, a node feature for the GCN includes a climate change impact indicator representing an impact of a climate change over a period on an entity represented by the node, the first set of edges represents online searches performed between the entities, and the second set of edges represents a correlation of equity returns between the entities,
wherein an ed

Assignees

• IBM
• Rensselaer Polytech Inst

Inventors

• Zhu Yada
• Liu Sijia
• Gupta Aparna
• Palepu Sai Radhakrishna Manikant Sarma
• Kar Koushik
• Popa Lucian
• Bhaskaran Kumar
• Gaur Nitin

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06Q40/08

  Insurance · CPC title

• G06Q40/03

  Credit; Loans; Processing thereof · CPC title

• G06Q10/0635

  Risk analysis of enterprise or organisation activities · CPC title

• G06Q30/0201

  Market modelling; Market analysis; Collecting market data · CPC title