Auto-discovery of reasoning knowledge graphs in supply chains

Patent metadata

Field	Value
Publication number	US-11928699-B2
Application number	US-202117219272-A
Country	US
Kind code	B2
Filing date	Mar 31, 2021
Priority date	Mar 31, 2021
Publication date	Mar 12, 2024
Grant date	Mar 12, 2024

Abstract

Methods, systems, and computer program products for auto-discovery of reasoning knowledge graphs in supply chains are provided herein. A computer-implemented method includes obtaining a spatiotemporal query related to a demand of at least one product in a supply chain; analyzing the spatiotemporal query to identify one or more parameters affecting the demand of the at least one product, wherein the one or more parameters comprise at least one of one or more climate parameters and one or more disruptive event parameters; generating a knowledge graph comprising information indicating an impact on the demand of the at least one product for at least a portion of the one or more parameters; and outputting, to a user interface, an explanation of a predicted demand forecast for the at least one product based at least in part on the knowledge graph.

First claim

What is claimed is:

1. A computer-implemented method, the method comprising:
obtaining a natural language spatiotemporal query related to a demand of at least one product in a supply chain;
applying one or more natural language processes to the natural language spatiotemporal query to identify a context of the natural language spatiotemporal query;
determining, based at least in part on the context, one or more parameters affecting the demand of the at least one product, wherein the one or more parameters comprise at least one of one or more climate parameters and one or more disruptive event parameters;
generating a set of counterfactual queries by perturbing the natural language spatiotemporal query with respect to at least one of the one or more parameters, wherein the generating comprises traversing a search space corresponding to the one or more parameters using an iterative search-based optimization algorithm;
computing one or more confidence values for one or more corresponding demand forecasts generated by a prediction model for the set of counterfactual queries;
constructing a knowledge graph comprising a set of nodes for at least a portion of the set of counterfactual queries, wherein each node in the set comprises information indicating an impact on the demand of the at least one product for at least a portion of the one or more parameters, and wherein the constructing comprises pruning one or more nodes in the set in response to determining that at least a first one of the computed confidence values satisfies a confidence value threshold corresponding to the computed first confidence value;
determining that at least a second one of the computed confidence values does not satisfy a confidence value threshold corresponding to the at least one computed second confidence value;
in response to determining that the at least second one of the computed confidence values does not satisfy the confidence value threshold corresponding to the at least one computed second confidence value, automatically initiating an interactive conversation with the user, wherein the interactive conversation comprises:
generating and outputting at least one question to the user, wherein the at least one question is generated based at least in part on the context; and
updating the at least one second computed confidence value based on feedback received from the user for the at least one question; and
outputting, to a user interface, an explanation of a predicted demand forecast for the at least one product based at least in part on the knowledge graph;

wherein the method is carried out by at least one computing device.

2. The computer-implemented method of claim 1 , wherein said determining comprises:
identifying one or more entities and one or more domain ontologies for at least one particular stage in the supply chain that is related to the natural language spatiotemporal query.

3. The computer-implemented method of claim 1 , comprising:
determining, using an explainable model, the impact of each of the one or more parameters on the demand forecasts generated by the prediction model.

4. The computer-implemented method of claim 1 , wherein:
each counterfactual query in the set corresponds to a different value or different range of values for a given one of the parameters; and
the generating the set of counterfactual queries comprises filtering the set of counterfactual queries based on at least one of: one or more geographical constraints related to the natural language spatiotemporal query and one or more domain-specific constraints related to the natural language spatiotemporal query.

5. The method of claim 4 , wherein:
the set of nodes comprises entity nodes capturing different scenarios corresponding to at least a portion of the one or more parameters; and
the knowledge graph comprises edges indicating relationships between pairs of the entity nodes, wherein the relationships correspond to at least one of: a predicted change in demand, a model confidence score, and a model uncertainty score.

6. The computer-implemented method of claim 1 , wherein the context comprises at least one of: one or more characteristics of the at least one product, a region related to the natural language spatiotemporal query, a particular stage within the supply chain related to the natural language spatiotemporal query.

7. The computer-implemented method of claim 1 , wherein the one or more climate parameters correspond to at least one of temperature and rainfall.

8. The computer-implemented method of claim 1 , wherein the one or more disruptive event parameters correspond to at least one of: a manmade disaster, a natural disaster, and an environmental disaster.

9. The computer-implemented method of claim 1 , comprising:
causing, based on the explanation, initiation of at least one proactive action to at least minimize the impact on the demand of the at least one product for at least a portion of the one or more parameters.

10. The computer-implemented method of claim 2 , wherein identifying the one or more entities and the one or more domain ontologies for the at least one particular stage in the supply chain comprises:
generate at least one vector representation for the natural language spatiotemporal query using at least one transformer-based machine learning model; and
processing the generated at least one vector representation with at least one classification model that is configured to determine the at least one particular stage of the supply chain associated with the natural language spatiotemporal query.

11. A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computing device to cause the computing device to:
obtain a natural language spatiotemporal query related to a demand of at least one product in a supply chain;
apply one or more natural language processes to the natural language spatiotemporal query to identify a context of the natural language spatiotemporal query;
determine, based at least in part on the context, one or more parameters affecting the demand of the at least one product, wherein the one or more parameters comprise at least one of one or more climate parameters and one or more disruptive event parameters;
generate a set of counterfactual queries by perturbing the natural language spatiotemporal query with respect to at least one of the one or more parameters, wherein the generating comprises traversing a search space corresponding to the one or more parameters using an iterative search-based optimization algorithm;
compute one or more confidence values for one or more corresponding demand forecasts generated by a prediction model for the set of counterfactual queries;
construct a knowledge graph comprising a set of nodes for at least a portion of the set of counterfactual queries, wherein each node in the set comprises information indicating an impact on the demand of the at least one product for at least a portion of the one or more parameters, and wherein the constructing comprises pruning one or more nodes in the set in response to determining that a given computed confidence value satisfies a corresponding confidence threshold;
in response to determining that at least one of the computed confidence values does not satisfy the corresponding confidence value threshold, automatically initiate an interactive conversation with the user, wherein the interactive conversation comprises:
generate and output at least one question to the user, wherein the at least one question is generated based at least in part on the context; and
update the at least one computed confide

Assignees

• IBM

Inventors

• MARVANIYA Smitkumar Narotambhai
• Guruprasad Ranjini Bangalore
• Godbole Shantanu R
• Kulkarni Kedar
• Singh Jitendra
• De Mel Geeth Ranmal
• Tomsett Richard J
• Weldemariam Komminist

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06Q30/0205 Primary

  based on location or geographical consideration · CPC title

• G06F16/90332

  Natural language query formulation or dialogue systems · CPC title

• G06F16/9537

  Spatial or temporal dependent retrieval, e.g. spatiotemporal queries · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title