Using graph queries to obtain results from machine learning models

Patent metadata

Field	Value
Publication number	US-12072936-B2
Application number	US-202117313769-A
Country	US
Kind code	B2
Filing date	May 6, 2021
Priority date	Dec 8, 2020
Publication date	Aug 27, 2024
Grant date	Aug 27, 2024

Abstract

A computer-implemented method performed at least in part by a graph database configured to store at least one graph dataset. The method includes receiving a graph query configured to be performed against a machine learning model, and communicating the graph query with a machine learning system that is configured to use the machine learning model to obtain model inference results and communicate those model inference results to the graph database application. The graph database provides query results based at least in part on the model inference results to an entity.

First claim

What is claimed is:

1. A computer-implemented method, comprising:
obtaining, at a graph database, a graph query that identifies criteria to identify a graph data structure responsive to the graph query, wherein the graph query identifies input data to be used by a trained machine learning model to generate at least a portion of the graph data structure;
executing the trained machine learning model against the input data to obtain an inference result representing a first subset of graph data;
applying the graph query to the first subset of the graph data represented by the inference result to generate first query results corresponding to a first portion of the graph data structure that is responsive to the graph query;
applying the graph query to a second subset of the graph data stored in the graph database to generate second query results corresponding to a second portion of the graph data structure that is responsive to the graph query;
combining at least the first portion of the graph data structure generated by application of the graph query to the first subset of the graph data represented by the inference result and the second portion of the graph data structure generated by application of the graph query to the second subset of the graph data stored in the graph database to result in the graph data structure responsive to the graph query; and
providing, by the graph database the graph data structure in response to the graph query.

2. The computer-implemented method of claim 1 , further comprising:
exporting, by the graph database, the input data to be used by the trained machine learning model to generate the first portion of the graph data structure.

3. The computer-implemented method of claim 1 , further comprising:
using an application programing interface (“API”) to execute the trained machine learning model against the input data.

4. The computer-implemented method of claim 1 , further comprising:
communicating, by the graph database, at least one query to a search service to obtain external results corresponding to a third portion of the graph data structure, wherein the first, second, and third portions of the graph data structure are combined to result in the graph data structure responsive to the graph query.

5. The computer-implemented method of claim 1 , wherein the trained machine learning model was trained using at least a first portion of one or more graph datasets stored by the graph database, the inference result is a first model inference result, and the computer-implemented method further comprises:
updating at least one of the one or more graph datasets stored by the graph database to obtain one or more updated graph datasets;
exporting at least a second portion of the one or more updated graph datasets from the graph database; and
training the trained machine learning model using the second portion before the trained machine learning model is used to obtain a second model inference result.

6. A system, comprising:
one or more processors; and
memory that stores computer-executable instructions that are executable by the one or more processors to cause the system to implement:
a graph database to:
obtain a graph query to comprise criteria to identify a graph data structure responsive to the graph query
use a trained machine learning model to obtain an inference result;
use a first portion of the criteria to query the inference result to generate first query results corresponding to a first portion of the graph data structure;
use a second portion of the criteria to query graph data stored in the graph database to obtain second query results corresponding to a second portion of the graph data structure;
obtain the graph data structure by combining at least the first portion of the graph data structure and the second portion of the graph data structure; and
provide the graph data structure in response to the graph query.

7. The system of claim 6 , wherein the graph database is to:
cause a search service to obtain external third query results corresponding to a third portion of the graph data structure from at least one external data source based at least in part on at least one external query portion of the graph query, and transmit the third query results to the graph database, wherein the graph database is to obtain the graph data structure by combining at least the first, second, and third portions of the graph data structure.

8. The system of claim 6 , wherein the graph database is to export input graph data to be used by the trained machine learning model to generate the first portion of the graph data structure.

9. The system of claim 6 , wherein the graph database is to:
export at least a portion of a graph dataset stored by the graph database,
instruct a machine learning service to generate and train a set of candidate machine learning models, and
select one of the set of candidate machine learning models as the trained machine learning model before the trained machine learning model is caused to obtain the inference result.

10. The system of claim 6 , wherein the graph database is to instruct a machine learning service to:
obtain first node embeddings for a plurality of first nodes in an input graph dataset, a second node embedding for a second node in the input graph dataset, and a relationship embedding for a relationship type in the input graph dataset between the second node and each of the plurality of first nodes,
calculate a predicted first node embedding as a function of the second node embedding and the relationship embedding,
identify one or more of the first node embeddings that are most similar to the predicted first node embedding as being at least one predicted first node embedding, and
for each of the at least one predicted first node embedding, identifying one of the plurality of first nodes as a predicted first node.

11. The system of claim 10 , wherein the machine learning service uses a first scoring function to determine which of the first node embeddings are most similar to the predicted first node embedding, and the graph database is to instruct the machine learning service to:
obtain second node embeddings for a plurality of second nodes in the input graph dataset, and a first node embedding for a first node in the input graph dataset,
calculate a predicted second node embedding as a function of the first node embedding and the relationship embedding,
identify one or more of the second node embeddings that are most similar to the predicted second node embedding as being at least one predicted second node embeddings, the machine learning service using a second scoring function to determine which of the second node embeddings are most similar to the predicted second node embedding, the second scoring function being different from the first scoring function, and
for each of the at least one predicted second node embeddings, identifying one of the plurality of second nodes as a predicted second node.

12. The system of claim 11 , wherein the first scoring function is a DistMult scoring function, a TransE scoring function, or a RotatE scoring function.

13. One or more non-transitory computer-readable storage media storing thereon executable instructions that, as a result of being executed by one or more processors of a computer system, cause the computer system to implement a graph database application to:
store at least one graph dataset;
obtain a graph query to identify a trained machine learning model;
cause the trained machine learning model to be executed against a portion of graph data stored in the at least one graph dataset to obtain an inference resul

Assignees

• Amazon Tech Inc

Inventors

• Bharathy Karthik Gurumoorthy Subramanya
• Gupta Ankit
• Karypis George
• Vel Divyakala
• Bebee Bradley
• Adeshina Adesoji
• Song Xiang

Classifications

• G06N3/0985

  Hyperparameter optimisation; Meta-learning; Learning-to-learn · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06F16/9038

  Presentation of query results · CPC title

• G06N3/08 Primary

  Learning methods · CPC title