Granted Patent B2

US 12,222,988 · App. 18/359,790 · Granted Feb 11, 2025

Inventors: Steven Hicklin (York, GB); Noushin Rezapour Asheghi (Atlanta, GA)

Assignee: EQUIFAX INC.

G06F16/90344G06F16/9035G06F16/906G06F40/30G06N20/00

Abstract

Claims (90)

1. A computing system comprising:

a client device configured to transmit a search query regarding an entity to a query system;

a search computing device configured to:

search a digital dataset based on keywords to determine a set of digital content, each digital content item in the set of digital content including one or more keywords identified from the search query, and

transmit the set of digital content to the query system; and

the query system communicatively coupled, via a data network, to the client device and the search computing device, the query system positioned remotely from the client device and between the client device and the search computing device, the query system comprising:

a non-transitory computer-readable medium storing program code; and

one or more processing devices communicatively coupled to the non-transitory computer-readable medium, the one or more processing devices configured to execute the program code to perform operations comprising:

receiving, from the client device, the query regarding the entity, wherein the query includes a request for documents with a certain sentiment regarding the entity,

creating a result set of digital content from the set of digital content, wherein creating the result set of digital content comprises:

obtaining, from the search computing device, the set of digital content,

generating, based on one or more machine-learning models, a reduced set of digital content by removing non-core content and duplicates from the set of digital content,

executing a content-attribute machine-learning model for each digital content item in the reduced set of digital content to determine a sentiment score based on a respective portion of each digital content item in the set of digital content, and

selecting, as the result set of digital content, a subset of digital content from the set of digital content, wherein the subset of digital content is selected based on the sentiment score determined for each digital content item in the set of digital content being above a threshold value associated with the certain sentiment, and

transmitting, to the client device, a message configured for providing the client device with access to the result set of digital content.

2. The computing system of claim 1 , wherein obtaining the set of digital content comprises:

obtaining, from the query, a name of the entity;

matching the name of the entity to names of individuals associated with the entity;

identifying a user preference specified via input from the client device;

determining the one or more keywords based on the name of the entity, the names of individuals associated with the entity, and the user preference;

transmitting, to the search computing device, a keyword query having the one or more keywords; and

receiving the set of digital content from the search computing device that match the one or more keywords.

3. The computing system of claim 1 , wherein executing the content-attribute machine-learning model for each digital content item comprises:

performing a sentiment analysis on a portion of each digital content item to determine a sentiment of the portion; and

determining, for each digital content item in the set of digital content, the sentiment score based on the sentiment of the portion and a location of the one or more keywords in the digital content item.

4. The computing system of claim 1 , wherein the one or more machine-learning models comprise a classification machine-learning model and wherein generating the reduced set of digital content further comprises:

executing the classification machine learning model for the set of digital content to extract core content from each digital content item in the set of digital content, wherein the classification machine learning model comprises a neural network model that comprises one or more hidden layers of interconnected nodes and is configured to identify core content from the digital content item using features comprising one or more of lexical features, text statistics, or relative position of text blocks; and

removing, from the set of digital content, content identified as non-core content.

5. The computing system of claim 4 , wherein extracting the core content from each digital content item comprises:

applying a machine-learning model to each digital content item that identifies the core content and non-core content in each digital content item; and

removing, from each digital content item, the non-core content.

6. The computing system of claim 4 , wherein determining the sentiment score comprises applying the content-attribute machine-learning model to the core content of each digital content item of the set of digital content.

7. The computing system of claim 1 , wherein generating the reduced set of digital content comprises:

tokenizing (i) a first digital content item in the set of digital content and (ii) a second digital content item in the set of digital content,

determining an amount of duplicate data between the tokenized first digital content item in the set of digital content and the tokenized second digital content item in the set of digital content, and

modifying the set of digital content by removing the first digital content item from the set of digital content based on the amount of duplicate data exceeding a threshold amount.

8. A method comprising:

receiving, by a content-extraction system and from a client device, a query regarding an entity, the content-extraction system positioned remotely from the client device and between the client device and a search computing device that is configured to search a digital dataset based on keywords, wherein the query includes a request for documents with a certain sentiment regarding the entity;

creating, by the content-extraction system, a result set of digital content responsive to the query, wherein creating the result set of digital content comprises:

obtaining, from the search computing device, a set of digital content, each digital content item in the set of digital content including one or more keywords identified from the query,

generating, based on one or more machine-learning models, a reduced set of digital content by removing non-core content and duplicates from the set of digital content,

executing, a content-attribute machine-learning model for each digital content item in the reduced set of digital content to determine a sentiment score based on a respective portion of each digital content item in the reduced set of digital content, and

selecting, as the result set of digital content, a subset of digital content from the reduced set of digital content, wherein the subset of digital content is selected based on the sentiment score determined for each digital content item in the reduced set of digital content; and

transmitting, by the content-extraction system, the result set of digital content to the client device.

9. The method of claim 8 , wherein obtaining the set of digital content comprises:

obtaining, from the query, a name of the entity;

matching the name of the entity to names of individuals associated with the entity;

identifying a user preference specified via input received by the content-extraction system from the client device;

determining the one or more keywords based on the name of the entity, the names of individuals associated with the entity, and the user preference;

transmitting, to the search computing device, a keyword query having the one or more keywords; and

receiving the set of digital content from the search computing device that match the one or more keywords.

10. The method of claim 8 , wherein executing the content-attribute machine-learning model for each digital content item comprises:

performing a sentiment analysis on a portion of each digital content item to determine a sentiment of the portion; and

determining, for each digital content item in the set of digital content, a sentiment score based on the sentiment of the portion and a location of the one or more keywords in the digital content item.

11. The method of claim 8 , wherein the one or more machine-learning models comprises a classification machine learning model and wherein generating the reduced set of digital content comprises:

executing the classification machine-learning model for the set of digital content to extract core content from each digital content item in the set of digital content, wherein the classification machine-learning model comprises a neural network model that comprises one or more hidden layers of interconnected nodes and is configured to identify core content from the digital content item using features comprising one or more of lexical features, text statistics, or relative position of text blocks; and

removing, from the set of digital content, content identified as non-core content.

12. The method of claim 11 , wherein extracting the core content from each digital content item comprises:

applying a machine-learning model to each digital content item that identifies the core content and non-core content in each digital content item; and

removing, from each digital content item, the non-core content.

13. The method of claim 11 , wherein determining the sentiment score comprises applying the content-attribute machine-learning model to the core content of each digital content item of the set of digital content.

14. The method of claim 8 , wherein generating the reduced set of digital content comprises:

tokenizing (i) a first digital content item in the set of digital content and (ii) a second digital content item in the set of digital content,

determining an amount of duplicate data between the tokenized first digital content item in the set of digital content and the tokenized second digital content item in the set of digital content, and

modifying the set of digital content by removing the first digital content item from the set of digital content based on the amount of duplicate data exceeding a threshold amount.

15. A non-transitory computer-readable medium having program code stored thereon, wherein the program code, when executed by one or more processing devices of a filtering system, configures the filtering system to perform operations comprising:

receiving, from a client device, a query regarding an entity, wherein the filtering system positioned remotely from the client device and between the client device and a search computing device that is configured to search a digital dataset based on keywords and wherein the query includes a request for documents with a certain sentiment regarding the entity;

creating a result set of digital content responsive to the query, wherein creating the result set of digital content comprises:

obtaining, from the search computing device, a set of digital content, each digital content item in the set of digital content including one or more keywords identified from the query,

generating, based on one or more machine-learning models, a reduced set of digital content by removing non-core content and duplicates from the set of digital content,

executing a content-attribute machine-learning model for each digital content item in the reduced set of digital content to determine a sentiment score based on a respective portion of each digital content item in the reduced set of digital content, and

selecting, as the result set of digital content, a subset of digital content from the reduced set of digital content, wherein the subset of digital content is selected based on the sentiment score determined for each digital content item in the reduced set of digital content; and

transmitting to the client device the result set of digital content.

16. The non-transitory computer-readable medium of claim 15 , wherein obtaining the set of digital content comprises:

obtaining, from the query, a name of the entity;

matching the name of the entity to names of individuals associated with the entity;

identifying a user preference specified via input received from the client device;

determining the one or more keywords based on the name of the entity, the names of individuals associated with the entity, and the user preference;

transmitting, to the search computing device, a keyword query having the one or more keywords; and

receiving the set of digital content from the search computing device that match the one or more keywords.

17. The non-transitory computer-readable medium of claim 15 , wherein executing the content-attribute machine-learning model for each digital content item comprises:

performing a sentiment analysis on a portion of each digital content item to determine a sentiment of the portion; and

determining, for each digital content item in the set of digital content, a respective value of the sentiment score based on the sentiment of the portion and a location of the one or more keywords in the digital content item.

18. The non-transitory computer-readable medium of claim 15 , wherein the one or more machine-learning models comprises a classification machine-learning model and wherein generating the reduced set of digital content comprises:

executing the classification machine-learning model for the set of digital content to extract core content from each digital content item in the set of digital content, wherein the classification machine-learning model comprises a neural network model that comprises one or more hidden layers of interconnected nodes and is configured to identify core content from the digital content item using features comprising one or more of lexical features, text statistics, or relative position of text blocks; and

removing, from the set of digital content, content identified as non-core content.

19. The non-transitory computer-readable medium of claim 18 , wherein extracting the core content from each digital content item comprises:

applying a machine-learning model to each digital content item that identifies the core content and non-core content in each digital content item; and

removing, from each digital content item, the non-core content.

20. The non-transitory computer-readable medium of claim 18 , wherein determining the sentiment score comprises applying the content-attribute machine-learning model to the core content of each digital content item of the set of digital content.

Assignments (1)

Continuity (3)

References Cited (58)