Granted Patent B2

US 11,188,830 · App. 15/057,487 · Granted Nov 30, 2021

Inventors: Xiang Li (New York, NY); Sungjin Lee (Edgewater, NJ); Aasish Pappu (New York, NY); Yifan Hu (Mountain Lakes, NJ); Amanda Stent (New York, NY)

Assignee: VERIZON MEDIA INC.

G06N5/022G06F16/9035G06N20/00G06Q10/04G06Q30/0631H04L67/10H04L67/306

Abstract

Claims (83)

1. A method implemented on a computing device having at least one processor, storage, and a communication platform connected to a network for generating user profiles with semantic knowledge, the method comprising:

obtaining first information associated with activities of a user;

identifying one or more entities from the first information;

generating a list of augmented entities from the one or more entities based on second information, wherein the list of augmented entities includes the one or more entities, one or more inferred entities, and information indicating semantic relationships between the one or more entities and the one or more inferred entities, wherein each of the one or more inferred entities is assigned an initial weight, the initial weight being a weight of zero;

updating, for each of the one or more inferred entities, the initial weight associated thereto with an updated weight determined based on the semantic relationships between the one or more entities and the one or more inferred entities, wherein the updated weight is estimated using a latent factor model trained with user interaction data;

clustering the list of augmented entities into a set of hierarchical clusters based on (i) the semantic relationships between the one or more entities and the one or more inferred entities, and (ii) the updated weight for each of the one or more inferred entities; and

generating a set of profiles for the user based on the set of hierarchical clusters, wherein each profile is to be used to personalize content recommendations for the user.

2. The method of claim 1 , wherein generating the list of augmented entities comprises:

identifying the one or more inferred entities from the second information, wherein the one or more inferred entities relate to the one or more entities; and

adding the one or more inferred entities to the one or more entities to generate the list of augmented entities,

wherein the second information comprises a knowledge archive of named entities.

3. The method of claim 2 , further comprising:

estimating user interests with respect to the list of augmented entities; and

incorporating the user interests with respect to the list of augmented entities into the set of profiles.

4. The method of claim 3 , wherein estimating the user interests with respect to the list of augmented entities further comprises:

identifying known user interests with respect to the one or more entities based on the information indicating the semantic relationships between the one or more entities and the one or more inferred entities;

estimating inferred user interests with respect to the one or more inferred entities based on the information indicating the semantic relationships between the one or more entities and the one or more inferred entities; and

estimating a strength of the user interests with respect to the list of augmented entities in accordance with the known user interests and the inferred user interests.

5. The method of claim 1 , further comprising:

receiving an input related to a user activity;

generating one or more local user profiles based on the user activity and the information indicating the semantic relationships between the one or more entities and the one or more inferred entities;

determining comparison results by comparing the one or more local user profiles with at least one profile of the set of profiles of the user; and

generating at least one updated profile of the set of profiles of the user by updating at least one profile of the set of profiles of the user with the one or more local user profiles, wherein the at least one updated profile is to be used to personalize the content recommendations for the user.

6. The method of claim 1 , wherein the one or more entities are identified from the first information based on:

an exact match or a similarity match between a first entity of the one or more entities and a first known entity from one or more known entities in a knowledge archive; or

a semantic relationship between a second entity of the one or more entities and a second known entity from the one or more known entities in the knowledge archive.

7. A system having at least one processor, storage, and a communication platform for generating user profiles with semantic knowledge, the system comprising:

a user activity analyzer, implemented by the at least one processor, configured to obtain first information associated with activities of a user;

an entity extractor, implemented by the at least one processor, configured to identify one or more entities from the first information;

an entity augmenting module, implemented by the at least one processor, configured to generate a list of augmented entities from the one or more entities based on second information, wherein the list of augmented entities includes the one or more entities, one or more inferred entities, and information indicating semantic relationships between the one or more entities and the one or more inferred entities, wherein each of the one or more inferred entities is assigned an initial weight, the initial weight being a weight of zero;

an entity clustering module, implemented by the at least one processor, configured to:

update, for each of the one or more inferred entities, the initial weight associated thereto with an updated weight determined based on the semantic relationships between the one or more entities and the one or more inferred entities, wherein the updated weight is estimated using a latent factor model trained with user interaction data, and

cluster the list of augmented entities into a set of hierarchical clusters based on (i) the semantic relationships between the one or more entities and the one or more inferred entities, and (ii) the updated weight for each of the one or more inferred entities; and

a user profile generating module, implemented by the at least one processor, configured to generate a set of profiles for the user based on the set of hierarchical clusters, wherein each profile of the set of profiles is to be used to personalize content recommendations for the user.

8. The system of claim 7 , wherein the entity augmenting module is further configured to:

identify the one or more inferred entities from the second information, wherein the one or more inferred entities relate to the one or more entities; and

add the one or more inferred entities to the one or more entities to generate the list of augmented entities,

wherein the second information includes semantic knowledge associated with the list of augmented entities.

9. The system of claim 8 , further comprising:

a user interest estimating module, implemented by the at least one processor, configured to:

estimate user interests with respect to the list of augmented entities; and

incorporate the user interests with respect to the list of augmented entities into the set of profiles.

10. The system of claim 9 , wherein the user interest estimating module configured to:

identify known user interests with respect to the one or more entities based on the information indicating the semantic relationships between the one or more entities and the one or more inferred entities;

estimate inferred user interests with respect to the one or more inferred entities based on the information indicating the semantic relationships between the one or more entities and the one or more inferred entities; and

estimate a strength of the user interests with respect to the list of augmented entities in accordance with the known user interests and the inferred user interests.

11. The system of claim 7 , wherein:

each hierarchical cluster of the set of hierarchical clusters defines a hierarchical relationship between the one or more entities and the one or more inferred entities of the list of augmented entities; and

each hierarchical cluster of the set of hierarchical clusters is generated using a clustering model with respect to different aspects of the semantic relationships between the one or more entities and the one or more inferred entities defined in the list of augmented entities.

12. A non-transitory machine-readable medium having information recorded thereon for generating profiles for a user, wherein the information, when read by the machine, effectuates operations comprising:

obtaining first information associated with activities of a user;

identifying one or more entities from the first information;

generating a list of augmented entities from the one or more entities based on second information, wherein the list of augmented entities includes the one or more entities, one or more inferred entities, and information indicating semantic relationships between the one or more entities and the one or more inferred entities, wherein each of the one or more inferred entities is assigned an initial weight, the initial weight being a weight of zero;

updating, for each of the one or more inferred entities, the initial weight associated thereto with an updated weight determined based on the semantic relationships between the one or more entities and the one or more inferred entities, wherein the updated weight is estimated using a latent factor model trained with user interaction data;

clustering the list of augmented entities into a set of hierarchical clusters based on (i) the semantic relationships between the one or more entities and the one or more inferred entities, and (ii) the updated weight for each of the one or more inferred entities; and

generating a set of profiles for the user based on the set of hierarchical clusters, wherein each profile of the set of profiles is to be used to personalize content recommendations for the user.

13. The medium of claim 12 , wherein generating the list of augmented entities comprises:

identifying the one or more inferred entities from the second information, wherein the one or more inferred entities relate to the one or more entities; and

adding the one or more inferred entities to the one or more entities to generate the list of augmented entities,

wherein the second information includes semantic knowledge associated with the list of augmented entities.

14. The medium of claim 13 , wherein the operations further comprise:

estimating user interests with respect to the list of augmented entities; and

incorporating the user interests with respect to the list of augmented entities into the set of profiles.

15. The medium of claim 14 , wherein estimating the user interests further comprises:

identifying known user interests with respect to the one or more entities based on the information indicating the semantic relationships between the one or more entities and the one or more inferred entities;

estimating inferred user interests with respect to the one or more inferred entities based on the information indicating the semantic relationships between the one or more entities and the one or more inferred entities; and

estimating a strength of the user interests with respect to the list of augmented entities in accordance with the known user interests and the inferred user interests.

16. The medium of claim 12 , wherein:

each hierarchical cluster of the set of hierarchical clusters defines a hierarchical relationship between the one or more entities and the one or more inferred entities of the list of augmented entities; and

each hierarchical cluster of the set of hierarchical clusters is generated using a clustering model with respect to different aspects of the semantic relationships between the one or more entities and the one or more inferred entities defined in the list of augmented entities.

17. A method implemented on a computing device having at least one processor, storage, and a communication platform connected to a network for recommending content using user profiling, the method comprising:

receiving an input from a user;

generating a set of user profiles based on information associated with activities of the user and a knowledge archive, wherein generating the set of user profiles comprises:

identifying one or more entities from the information,

generating a list of augmented entities from the one or more entities based on the knowledge archive, wherein the list of augmented entities includes the one or more entities, one or more inferred entities, and semantic relationships between the one or more entities and the one or more inferred entities, wherein each of the one or more inferred entities is assigned an initial weight, the initial weight being a weight of zero,

updating, for each of the one or more inferred entities, the initial weight associated thereto with an updated weight determined based on the semantic relationships between the one or more entities and the one or more inferred entities, wherein the updated weight is estimated using a latent factor model trained with user interaction data,

clustering the list of augmented entities into a set of hierarchical clusters based on (i) the semantic relationships between the one or more entities and one or more inferred entities, and (ii) the updated weight for each of the one or more inferred entities, wherein the set of user profiles is generated based on the set of hierarchical clusters;

generating a set of augmented user profiles by augmenting the set of user profiles based on a set of pre-constructed user profiles; and

recommending content to the user in response to the input based on the set of augmented user profiles,

wherein each of the set of augmented user profiles defines an aspect of user interests with respect to a plurality of entities and semantic relationships between the plurality of entities.

18. The method of claim 17 , wherein the set of user profiles comprises one or more local user profiles generated based on user activity and the semantic relationship between the one or more entities and the one or more inferred entities, generating the set of augmented user profiles comprises:

determining comparison results by comparing the one or more local user profiles with at least one pre-constructed user profile of the set of pre-constructed user profiles; and

generating one or more updated user profiles by updating the at least one pre-constructed user profile with the one or more local user profiles, wherein the set of augmented user profiles comprises the one or more updated user profiles.

Assignments (7)