Knowledge graphs in machine learning decision optimization

Patent metadata

Field	Value
Publication number	US-12536428-B2
Application number	US-202117183870-A
Country	US
Kind code	B2
Filing date	Feb 24, 2021
Priority date	Feb 24, 2021
Publication date	Jan 27, 2026
Grant date	Jan 27, 2026

Abstract

Training a machine learning model can include receiving time series data. A knowledge graph structure can be received including nodes and edges, the nodes representing entities associated with the time series data, the edges representing relationships between the nodes connected by the edges. A machine learning model can be structured to forecast a prediction using the time series data. The machine learning model can be structured to integrate the knowledge graph structure as an error term in the machine learning model. The machine learning model can be trained to forecast the prediction based on the time series data and the knowledge graph structure. The error term representing the knowledge graph structure can be regularized for sparsity during training.

First claim

What is claimed is:

1 . A computer-implemented method of training a machine learning prediction model, comprising:
receiving time series data representing time varying data of an asset; receiving a knowledge graph structure including nodes and edges, the nodes representing entities associated with the time series data, the edges representing relationships between the nodes connected by the edges, wherein the relationships represented by the edges are time varying and the knowledge graph structure provides additional information about the asset in addition to the time series data; structuring a machine learning model to forecast a prediction using the time series data, the machine learning model having a time series forecasting component and an error component, the time series forecasting component representing the time series data and the error component representing the additional information provided by the knowledge graph structure, the time series data and the knowledge graph structure having different types of co-evolving data-structures, wherein the entities and the relationships represented in the knowledge graph are integrated in the machine learning model as an error term in the error component; and training the machine learning model to forecast the prediction based on the time series data and the knowledge graph structure, the training including regularizing the error component for sparsity, the training including alternately updating weights of the machine learning model by proximal gradient descent and updating relationship representation of the knowledge graph structure via a conditional variance using a pseudo-likelihood method, during learning iterations of the training, the machine learning model jointly learning the time series data and the knowledge graph structure during the learning iterations, a connectivity information of the knowledge graph represented by the error component reducing over-parametrization in the time series forecasting component wherein at least some of the connectivity information of the knowledge graph represented in the error component are masked based on a symmetric weight matrix that measures strengths of the relationships represented in the knowledge graph.

2 . The method of claim 1 , wherein the machine learning model includes a neural network.

3 . The method of claim 1 , wherein the machine learning model includes a graph neural network.

4 . The method of claim 1 , wherein the knowledge graph structure is transformed to an inverse covariance matrix of a Gaussian error to be expressed as the error term in the machine learning model.

5 . The method of claim 4 , wherein the training jointly learns a prediction function of the machine learning model and the knowledge graph structure integrated as the error term represented by the inverse covariance matrix of the Gaussian error.

6 . The method of claim 1 , wherein the regularizing includes using a soft mask including a real value.

7 . The method of claim 1 , wherein the regularizing includes using a hard mask including a binary value.

8 . The method of claim 1 , wherein the prediction includes financial portfolio composition.

9 . The method of claim 1 , further including providing a user interface for allowing a user to configure a type of the machine learning model, the time series data and the knowledge graph structure.

10 . A system comprising:
a processor; a memory device coupled with the processor; the processor configured to at least:
receive time series data representing time varying data of an asset;
receive a knowledge graph structure including nodes and edges, the nodes representing entities associated with the time series data, the edges representing relationships between the nodes connected by the edges, wherein the relationships represented by the edges are time varying and the knowledge graph structure provides additional information about the asset in addition to the time series data;
structure a machine learning model to forecast a prediction using the time series data, the machine learning model having a time series forecasting component and an error component, the time series forecasting component representing the time series data and the error component representing the additional information provided by the knowledge graph structure, the time series data and the knowledge graph structure having different types of co-evolving data-structures, wherein the entities and the relationships represented in the knowledge graph are integrated in the machine learning model as an error term in the error component; and
train the machine learning model to forecast the prediction based on the time series data and the knowledge graph structure, the processor configured to regularize the error component for sparsity during training, the processor configured to train the machine learning model by alternately updating weights of the machine learning model by proximal gradient descent and updating relationship representation of the knowledge graph structure via a conditional variance using a pseudo-likelihood method, during learning iterations of the training, the machine learning model jointly learning the time series data and the knowledge graph structure during the learning iterations, a connectivity information of the knowledge graph represented by the error component reducing over-parametrization in the time series forecasting component wherein at least some of the connectivity information of the knowledge graph represented in the error component are masked based on a symmetric weight matrix that measures strengths of the relationships represented in the knowledge graph.

11 . The system of claim 10 , wherein the machine learning model includes a neural network.

12 . The system of claim 10 , wherein the machine learning model includes a graph neural network.

13 . The system of claim 10 , wherein the knowledge graph structure is transformed to an inverse covariance matrix of a Gaussian error to be expressed as the error term in the machine learning model.

14 . The system of claim 13 , wherein in training the machine learning model, the processor is configured to jointly learns a prediction function of the machine learning model and the knowledge graph structure integrated as the error term represented by the inverse covariance matrix of the Gaussian error.

15 . The system of claim 10 , wherein the processor configured to regularize the error term using a soft mask including a real value.

16 . The system of claim 10 , wherein the processor configured to regularize the error term using a hard mask including a binary value.

17 . The system of claim 10 , wherein the prediction includes financial portfolio composition.

18 . The system of claim 10 , further including a user interface configured to received from a user a selection of a type of the machine learning model, the time series data and the knowledge graph structure.

19 . A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions readable by a device to cause the device to:
receive time series data representing time varying data of an asset; receive a knowledge graph structure including nodes and edges, the nodes representing entities associated with the time series data, the edges representing relationships between the nodes connected by the edges, wherein the relationships represented by the edges are time varying and the knowledge graph structure provides additional

Assignees

• IBM
• Massachusetts Inst Technology

Inventors

• Zhu Yada
• Zhang Yang
• Chen Pin-Yu
• Mazumder Rahul
• Ibrahim Shibal
• Chen Wenyu

Classifications

• G06Q40/06 Primary

  Asset management; Financial planning or analysis · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/042

  Knowledge-based neural networks; Logical representations of neural networks · CPC title