Entity relation strength identification using spatiotemporal data

Patent metadata

Field	Value
Publication number	US-12412133-B2
Application number	US-202217655740-A
Country	US
Kind code	B2
Filing date	Mar 21, 2022
Priority date	Mar 21, 2022
Publication date	Sep 9, 2025
Grant date	Sep 9, 2025

Abstract

A method includes obtaining information associated with multiple collocation events. Each collocation event is associated with an occurrence where multiple entities were collocated with one another. The information associated with each collocation event identifies the entities associated with the collocation event, a duration of the collocation event, one or more locations associated with the collocation event, and a time associated with the collocation event. The method also includes identifying strengths of different relationships involving different ones of the entities based on the information. The strengths are determined using one or more features of the collocation events.

First claim

What is claimed is:

1. A method comprising:
obtaining information associated with multiple collocation events from multiple data sources, wherein each collocation event is associated with an occurrence where multiple entities were collocated with one another;
pre-processing the information from the multiple data sources into a unified data format using a first machine learning model, the unified data format defining:
the multiple entities associated with a collocation event;
a duration of the collocation event;
one or more locations associated with the collocation event; and
a time associated with the collocation event;

determining, by a second machine learning model, multiple features of the multiple collocation events using the information in the unified data format, wherein the second machine learning model is distinct from the first machine learning model;
providing the multiple determined features of the multiple collocation events to a third machine learning model, wherein the third machine learning model is distinct from the first machine learning model and the second machine learning model; and
identifying strengths of different relationships involving different ones of the multiple entities using the third machine learning model based on the multiple determined features of the multiple collocation events;
wherein the third machine learning model is trained to identify, using the information in the unified data format, the strengths of the different relationships by generating scores based on the multiple determined features of the multiple collocation events.

2. The method of claim 1 , wherein the multiple determined features of the multiple collocation events comprise:
one or more count features associated with one or more frequencies or numbers of collocation events involving specified entities of the multiple entities;
one or more distance features associated with one or more distances between or among the specified entities during the multiple collocation events; and
one or more duration features associated with one or more lengths of collocation involving the specified entities during the multiple collocation events.

3. The method of claim 2 , wherein:
the multiple determined features of the multiple collocation events are determined for each of multiple timeframes; and
the strengths of the different relationships are determined using the multiple determined features of the multiple collocation events during the multiple timeframes.

4. The method of claim 1 , further comprising:
scaling feature values of at least one of the multiple determined features to generate scaled features of the at least one of the multiple determined features;
wherein the strengths of the different relationships are determined using the scaled feature values of the at least one of the multiple determined features.

5. The method of claim 4 , wherein the scaled feature values of all of the determined features span a common range of values.

6. The method of claim 4 , wherein:
a first determined feature of the multiple determined features uses values below a first threshold to identify weaker associations involving the multiple entities and values above the first threshold to identify stronger associations involving the multiple entities, wherein a strength of an association reflects a quantification of a grouping of entities of the multiple entities;
a second determined feature of the multiple determined features uses values below a second threshold to identify the stronger associations involving the multiple entities and values above the second threshold to identify the weaker associations involving the multiple entities; and
scaling the feature values comprises scaling the feature values of the first determined feature or the second determined feature to reverse values representing the stronger associations involving the multiple entities and the weaker associations involving the multiple entities.

7. The method of claim 1 , wherein identifying the strengths of the different relationships comprises, for each of the different relationships, determining an aggregation of feature values associated with entities among multiple entities involved in a relationship.

8. The method of claim 1 , wherein the second machine learning model is trained during unsupervised learning to perform anomaly detection.

9. The method of claim 1 , wherein identifying the strengths of the different relationships comprises using one or more statistical calculations to identify the strengths of the different relationships.

10. The method of claim 1 , wherein the second machine learning model is trained during semi-supervised or supervised learning using labeled training data.

11. The method of claim 1 , wherein:
identifying the strengths of the different relationships comprises, for each of the different relationships, determining an aggregation of feature values associated with multiple entities involved in a relationship; and
unequal weights are associated with feature values of each determined feature of the multiple collocation events.

12. The method of claim 11 , wherein:
the multiple collocation events comprise meetings between a plurality of entities at a specific location;
the feature values include feature values associated with a frequency at which the meetings at the specific location occur;
the feature values associated with the frequency at which the meetings at the specific location occur have unequal weights; and
the unequal weights are based on different distances that different entities of the plurality of entities traveled to the specific location.

13. The method of claim 1 , wherein at least one of the multiple determined features is associated with distances traveled by one or more of the multiple entities to or from one or more of the multiple collocation events.

14. An apparatus comprising:
at least one processing device configured to:
obtain information associated with multiple collocation events from multiple data sources, wherein each collocation event is associated with an occurrence where multiple entities were collocated with one another;
pre-process the information from the multiple data sources into a unified data format using a first machine learning model, the unified data format defining:
the multiple entities associated with a collocation event;
a duration of the collocation event;
one or more locations associated with the collocation event; and
a time associated with the collocation event;

determine, by a second machine learning model, multiple features of the multiple collocation events using the information in the unified data format, wherein the second machine learning model is distinct from the first machine learning model;
provide the multiple determined features of the multiple collocation events to a third machine learning model, wherein the third machine learning model is distinct from the first machine learning model and the second machine learning model; and
identify strengths of different relationships involving different ones of the multiple entities using the third machine learning model based on the multiple determined features of the multiple collocation events;

wherein the third machine learning model is trained to identify, using the information in the unified data format, the strengths of the different relationships by generating scores based on the multiple determined features of the multiple collocation events.

15. The apparatus of claim 14 , wherein the multiple determined features of the multiple collocation events comprise:
one or more count feat

Assignees

• C3 Ai Inc

Inventors

• Lin Chih-Hsu
• Ye Franklin Q
• Juban Romain F
• Anderson Bryan C

Classifications

• G06Q10/063 Primary

  Operations research, analysis or management · CPC title

• G06F16/9537

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

• G06F16/909

  using geographical or spatial information, e.g. location (spatiotemporally dependent retrieval from the web G06F16/9537) · CPC title