Granted Patent B2

US 12,321,355 · App. 18/382,038 · Granted Jun 3, 2025

Inventors: Gann Bierner (Oakland, CA); Robert Weis (Oakland, CA)

Assignee: Ancestry.com Operations Inc.

G06F16/24575G06F18/2415

Abstract

Claims (65)

1. A computer-implemented method for linking a natural-language user query at a genealogy online system to a specific preexisting function of the genealogy online system, the computer-implemented method comprising:

causing to display, at a graphical user interface associated with a genealogy online system, a search box, the genealogy online system configured to provide functions comprising family-tree building and historical record search;

receiving a query from a user entered at the search box;

using a machine learning language model to determine an intent of the user associated with the query, the machine learning language model being trained to receive a query and classify the query to an intent mappable to a plurality of functions of the genealogy online system, wherein the machine learning language model is a transformer model,

mapping the determined intent from the machine learning language model to a plurality of functions of a predetermined table of intents and associated functions of the genealogy online system,

outputting a prediction regarding the mapped plurality of functions of the genealogy online system based on the mapping of the determined intent of the user; and

causing to display, at the graphical user interface as a result of the query, one or more links to the predicted one or more functions of the genealogy online system based on the intent determined by the machine learning language model.

2. The computer-implemented method of claim 1 , wherein training of the machine learning language model comprises:

receiving a set of training samples that include queries input by users and intent labels corresponding to the queries;

converting the queries to embeddings;

inputting, in a forward propagation, the embeddings to the machine learning language model to predict intents;

comparing the predicted intents with the intent labels; and

backpropagating comparison results to the machine learning language model to adjust one or more parameters of the machine learning language model.

3. The computer-implemented method of claim 1 , further comprising:

converting the query from the user that is in natural language to a structured query using a second machine learning language model;

determining the intent of the query using the machine learning language model;

launching a function of the genealogy online system; and

automatically sending the structured query to the function to generate a query result.

4. The computer-implemented method of claim 1 , wherein the functions of the genealogy online system comprises: a family tree viewer, a genealogical record research interface, a user profile setting, and a genetic data analysis tool.

5. The computer-implemented method of claim 1 , wherein the search box is located in a first webpage domain and the one or more functions of the genealogy online system are associated with second webpage domains that are different from the first webpage domain.

6. The computer-implemented method of claim 1 , wherein the search box is located on a landing page of the genealogy online system.

7. The computer-implemented method of claim 1 , wherein the machine learning language model is provided by a third party and the genealogy online system communicates with the machine learning language model via an Application Programming Interface (API).

8. The computer-implemented method of claim 1 , wherein the machine learning language model is fine-tuned with domain-specific data of the genealogy online system.

9. The computer-implemented method of claim 1 , wherein a relative of the user is identified based on the determined intent of the natural-language user query from a stitched genealogical tree database.

10. The computer-implemented method of claim 1 , wherein the machine learning language model is further configured to:

extract a plurality of entities and corresponding relationships from the natural-language user query, and

pre-populate a person-search query of a records database of the genealogy online system.

11. The computer-implemented method of claim 1 , wherein the machine learning language model comprises two machine learning model layers,

wherein a first machine learning model layer of the machine learning language model detects the intent of the natural-language user query, and

wherein a second machine learning model layer of the machine learning language model determines that the natural-language user query pertains to a person, the second machine learning language model determining that the natural-language user query pertains to a person by extracting at least one entity from the natural-language user query.

12. The computer-implemented method of claim 11 , wherein:

the second machine learning language model layer is first utilized to determine that the natural-language user query pertains to the person, and

the first machine learning language model layer detects the intent of the natural-language user query at least in part based on the determination that the natural-language user query pertains to the person.

13. The computer-implemented method of claim 1 , further comprising:

updating the predetermined table of intents and associated functions of the genealogy online system based on an interaction of the user with the displayed one or more links.

14. The computer-implemented method of claim 1 , wherein the predicted one or more functions of the genealogy online system comprises a genetic data analysis tool.

15. The computer-implemented method of claim 1 , wherein the prediction regarding the mapped one or more functions of the genealogy online system is additionally based on profile data extracted from a user profile of the user.

16. A system for linking a natural-language user query at a genealogy online system to a specific preexisting function of the genealogy online system, the system comprising:

a machine learning language model; and

a genealogy online system to provide functions comprising family-tree building and historical record search, the genealogy online system comprising memory and one or more processors, the memory configured to store code comprising instructions, the instructions, when executed by the one or more processors, cause the one or more processors to perform steps comprising:

causing to display, at a graphical user interface, a search box for receiving a query from a user entered at the search box;

using the machine learning language model to determine an intent of the user associated with the query, the machine learning language model being trained to receive a query and classify the query to an intent mappable to a plurality of functions of the genealogy online system, wherein the machine learning language model is a transformer model,

mapping the determined intent from the machine learning language model to a plurality of functions of a predetermined table of intents and associated functions of the genealogy online system,

outputting a prediction regarding the mapped plurality of functions of the genealogy online system based on the mapping of the determined intent of the user; and

causing to display, at the graphical user interface as a result of the query, one or more links to one or more functions of the genealogy online system based on the intent determined by the machine learning language model.

17. The system of claim 16 , wherein training of the machine learning language model comprises:

receiving a set of training samples that include queries input by users and intent labels corresponding to the queries;

converting the queries to embeddings;

inputting, in a forward propagation, the embeddings to the machine learning language model to predict intents;

comparing the predicted intents with the intent labels; and

backpropagating comparison results to the machine learning language model to adjust one or more parameters of the machine learning language model.

18. A system for linking a natural-language user query at a genealogy online system to a specific preexisting function of the genealogy online system comprising:

a graphical user interface configured to display a search box for receiving a query from a user entered at the search box; and

a genealogy online system to provide functions comprising family-tree building and historical record search, the genealogy online system comprising memory and one or more processors, the memory configured to store code comprising instructions, the instructions, when executed by the one or more processors, cause the one or more processors to perform steps comprising:

using a machine learning language model to determine an intent of the user associated with the query, the machine learning language model being trained to receive a query and classify the query to an intent mappable to a plurality of functions of the genealogy online system, wherein the machine learning language model is a transformer model,

mapping the determined intent from the machine learning language model to a plurality of functions of a predetermined table of intents and associated functions of the genealogy online system,

outputting a prediction regarding the mapped plurality of functions of the genealogy online system based on the mapping of the determined intent of the user; and

causing to display, at the graphical user interface as a result of the query, one or more links to one or more functions of the genealogy online system based on the intent determined by the machine learning language model.

19. The system of claim 18 , wherein the steps further comprise:

converting the query from the user that is in natural language to a structured query using a second machine learning language model;

determining the intent of the query using the machine learning language model;

launching a function of the genealogy online system; and

automatically sending the structured query to the function to generate a query result.

20. The system of claim 18 , wherein the search box is located in a first webpage domain and the one or more functions of the genealogy online system are associated with second webpage domains that are different from the first webpage domain.

21. The system of claim 18 , wherein the machine learning language model is fine-tuned with domain-specific data of the genealogy online system.

Assignments (3)

Continuity (3)

References Cited (16)

Cited By (1)

US 12,675,644