Performing targeted searching based on a user profile

Patent metadata

Field	Value
Publication number	US-11921728-B2
Application number	US-202117163278-A
Country	US
Kind code	B2
Filing date	Jan 29, 2021
Priority date	Jan 29, 2021
Publication date	Mar 5, 2024
Grant date	Mar 5, 2024

Abstract

Aspects of the present disclosure relate to systems and methods for performing targeted searching based on a user profile. In examples, a user profile including a user embedding may be retrieved based on the receipt of a user indication. The user embedding may be created based on one or more user interest. A plurality of document embeddings may be identified based on the user embedding, where each document embedding of the plurality of document embeddings is determined to be within a first distance of the user embedding. In examples, a ranking for each document embedding of the plurality of document embeddings may be generated, where the ranking for each document embedding of the plurality of document embeddings is based on the user embedding. At least one document may be recommend based on a ranking associated with a document embedding.

First claim

What is claimed is:

1. A method for performing a targeted search based on user interests, the method comprising:
receiving a user identifier for a user;
retrieving a user embedding based on the user identifier, wherein the user embedding is generated using a machine learning model based on received indications of user characteristics, user interests, and user actions as a low-dimensional space representation of the user;
identifying a set of document embeddings from a plurality of document embeddings using the user embedding, each document embedding of the set of document embeddings determined to be within a first distance of the user embedding;
generating a ranking for each document embedding of the set of document embeddings based on the user embedding; and
recommending to the user at least one document based on the ranking for each document embedding of the set of document embeddings.

2. The method of claim 1 , wherein the user embedding further comprises a low-dimensional space representation of a translated higher-dimensional vector derived from information associated with the user.

3. The method of claim 2 , further comprising:
generating the ranking for each document embedding of the set of document embeddings using a machine learning model, the machine learning model ranking each document embedding of the set of document embeddings based on one or more of relevancy, novelty, and diversity.

4. The method of claim 1 ,
wherein the user interests further comprise an indication of one or more topics of interest to the user and one or more topics that are not of interest to the user, and
wherein the user actions further comprise one or more previous documents viewed by the user.

5. The method of claim 1 , further comprising:
extracting location and content information from a document;
providing the location and content information for the document to a machine learning model;
generating, via the machine learning model, a document embedding specific to the document; and
adding the document embedding to the plurality of document embeddings.

6. The method of claim 1 , further comprising causing the recommended at least one document to be displayed at an output device associated with the user.

7. The method of claim 6 , further comprising:
receiving an indication that a user selected the recommended at least one document displayed to the output device;
generating another user embedding based on the received indication;
identifying a second set of document embeddings from the plurality of document embeddings based on the another user embedding;
generating a ranking for each document embedding of the second set of document embeddings, wherein the ranking for each document embedding of the second set of document embeddings is based on the another user embedding; and
recommending to the user at least one document based on the ranking for each document embedding of the second set of document embeddings.

8. A system for performing a targeted search based on user interests, the system comprising:
a processor;
memory including instructions, which when executed by the processor, cause the processor to:
extract information from each document of a plurality of documents;
for each document of the plurality of documents, generate a document embedding based on the extracted information;
receive user characteristics, user interest information, and user actions for a user;
generate a user embedding using a machine learning model based on the received user characteristics, user interest information, and user actions, where the user embedding and the document embedding sharing the same semantic space, wherein the user embedding is a low-dimensional space representation of the user;
identify a set of document embeddings from the plurality of document embeddings using the user embedding, each document embedding of the set of document embeddings being within a first distance of the user embedding;
generate a ranking for each document embedding of the set of document embeddings based on the user embedding; and
recommend to the user, at least one document based on the document embedding ranking.

9. The system of claim 8 , wherein the recommended at least one document is rendered to a display device associated with a user associated with the user embedding.

10. The system of claim 8 , wherein the instructions, which when executed by the processor, cause the processor to:
receive an indication that the user selected the recommended at least one document;
generate another user embedding associated with the user based on the received indication;
identify a second set of document embeddings from the plurality of document embeddings based on the another user embedding;
generate a ranking for each document embedding of the second set of document embeddings, wherein the ranking for each document embedding of the second set of document embeddings is based on the another user embedding; and
recommend to the user at least one document based on the ranking for each document embedding of the second set of document embeddings.

11. The system of claim 10 , wherein the instructions, which when executed by the processor, cause the processor to:
receive a user identifier; and
retrieve a user profile based on the user identifier, the user profile including the user embedding.

12. The system of claim 10 , further comprising instructions, which when executed by the processor, cause the processor to:
receive, for a plurality of documents associated with the set of document embeddings, user interest information from a plurality of users; and
identify a subset of the set of document embeddings based on the user interest information from the plurality of users matching user interest information for the user,
wherein the recommended at least one document is based on the subset of the set of document embeddings.

13. The system of claim 8 , wherein the instructions, which when executed by the processor, cause the processor to generate the ranking for each document embedding of the plurality of document embeddings using a machine learning model, the machine learning model ranking each document embedding based on one or more of relevancy, novelty, and diversity.

14. The system of claim 13 , wherein the user embedding further comprises a low-dimensional space representation of a translated higher-dimensional vector derived from information associated with the user.

15. A computer storage medium including instructions, which when executed by a processor, cause the processor to:
receive a user identifier for a user;
retrieve a user profile based on the user identifier, the user profile including a user embedding generated using a machine learning model based on received indications of user characteristics, user interests, and user actions as a low-dimensional space representation of the user;
identify a plurality of document embeddings using the user embedding, each document embedding of the plurality of document embeddings determined to be within a first distance of the user embedding;
generate a ranking for each document embedding of the plurality of document embeddings based on the user embedding; and
recommend to the user at least one document based on the ranking for each document embedding of the plurality of document embeddings.

16. The computer storage medium of claim 15 , wherein the user embedding further comprises a low-dimensional space representation of a translated higher-dimensional vector derived from information associated with the user.

17. The computer sto

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Ahmed Junaid
• Malik Waleed
• Overwijk Arnold

Classifications

• G06F16/24578 Primary

  using ranking · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F16/9035 Primary

  Filtering based on additional data, e.g. user or group profiles · CPC title