Method and apparatus for processing web content, device, and computer storage medium

Patent metadata

Field	Value
Publication number	US-11714864-B2
Application number	US-202117359874-A
Country	US
Kind code	B2
Filing date	Jun 28, 2021
Priority date	May 28, 2019
Publication date	Aug 1, 2023
Grant date	Aug 1, 2023

Abstract

A method and an apparatus for processing web content, a device, and a computer storage medium are provided. A long-term feature group and a short-term feature group are determined from historical browsing data of a user according to generation time points of elements in the historical browsing data. A long-term encoding vector corresponding to the long-term feature group is determined according to similarities between elements in the long-term feature group, a user embedding vector corresponding to the short-term feature group is determined according to the long-term encoding vector and similarities between elements in the short-term feature group, and at least one web content is determined as a recommendation candidate and provided to the user.

First claim

What is claimed is:

1. A method for processing web content, performed by a server, the method comprising:
determining a long-term feature group including first elements that reflect a long-term browsing interest of a user, by using first historical browsing data of the user that are generated in a first predetermined period of a past time, the first elements including items related to a content of the first historical browsing data;
determining a short-term feature group including second elements that reflect a short-term browsing interest of the user, by using second historical browsing data of the user that are generated in a second predetermined period of the past time, the second predetermined period being shorter than the first predetermined period and relatively recent to a current time, the second elements including items related to a content of the second historical browsing data;
determining a long-term encoding vector that reflects similarities between the first elements in the long-term feature group;
determining a user embedding vector that reflects similarities between the long-term encoding vector and the second elements in the short-term feature group; and
determining, as a recommendation candidate, at least one web content based on a similarity between the at least one web content and the user embedding vector, and providing the at least one web content to the user.

2. The method according to claim 1 , wherein the long-term feature group comprises a plurality of long-term sub-type feature groups, and the determining the long-term encoding vector comprises:
determining a long-term embedding sub-vector corresponding to each long-term sub-type feature group according to similarities between first elements in each long-term sub-type feature group; and
determining the long-term encoding vector corresponding to the long-term feature group according to similarities between long-term embedding sub-vectors.

3. The method according to claim 1 , wherein the determining the long-term encoding vector comprises:
determining the similarities between the first elements in the long-term feature group according to an attention network model; and
determining the long-term encoding vector corresponding to the long-term feature group according to the similarities between the first elements in the long-term feature group.

4. The method according to claim 1 , wherein the short-term feature group comprises a plurality of short-term sub-type feature groups, and the determining the user embedding vector comprises:
determining short-term embedding sub-vectors corresponding to the plurality of short-term sub-type feature groups according to similarities between the long-term encoding vector and second elements in each short-term sub-type feature group; and
determining the user embedding vector corresponding to the short-term feature group according to similarities between the short-term embedding sub-vectors.

5. The method according to claim 1 , wherein the determining the user embedding vector comprises:
determining similarities between the long-term encoding vector and the second elements in the short-term feature group according to an attention network model; and
determining the user embedding vector corresponding to the short-term feature group according to the long-term encoding vector and the similarities between the second elements in the short-term feature group.

6. The method according to claim 1 , wherein the determining the at least one web content comprises:
determining a similarity between the at least one web content and the user embedding vector; and
determining, as the recommendation candidate, the at least one web content of which a similarity with the user embedding vector meets a preset condition.

7. The method according to claim 6 , wherein the at least one web content comprises a plurality of web contents, and the method further comprises:
ranking the plurality of web contents according to categories of the plurality of web contents; and
adjusting an order of displayed web contents according to a result of the ranking of the plurality of web contents.

8. An apparatus for processing web content, the apparatus comprising:
at least one memory configured to store program code; and
at least one processor configured to read the program code and operate as instructed by the program code, the program code comprising:
long-term feature group determining code configured to cause at least one of the at least one processor to determine a long-term feature group including first elements that reflect a long-term browsing interest of a user, by using first historical browsing data of the user that are generated in a first predetermined period of a past time, the first elements including items related to a content of the first historical browsing data;
short-term feature group determining code configured to cause at least one of the at least one processor to determine a short-term feature group including second elements that reflect a short-term browsing interest of the user, by using second historical browsing data of the user that are generated in a second predetermined period of the past time, the second predetermined period being shorter than the first predetermined period and relatively recent to a current time, the second elements including items related to a content of the second historical browsing data;
long-term encoding vector determining code configured to cause at least one of the at least one processor to determine a long-term encoding vector that reflects similarities between first elements in the long-term feature group;
user embedding vector determining code configured to cause at least one of the at least one processor to determine a user embedding vector that reflects similarities between the long-term encoding vector and the second elements in the short-term feature group; and
recommendation candidate determining code configured to cause at least one of the at least one processor to determine, as a recommendation candidate, at least one web content based on a similarity between the at least one web content and the user embedding vector, and provide the at least one web content to the user.

9. The apparatus according to claim 8 , wherein the long-term feature group comprises a plurality of long-term sub-type feature groups, and
the long-term encoding vector determining code is further configured to cause at least one of the at least one processor to determine a long-term embedding sub-vector corresponding to each long-term sub-type feature group according to similarities between first elements in each long-term sub-type feature group; and determine the long-term encoding vector corresponding to the long-term feature group according to similarities between long-term embedding sub-vectors.

10. The apparatus according to claim 8 , wherein the long-term encoding vector determining code is further configured to cause at least one of the at least one processor to determine the similarities between the first elements in the long-term feature group according to an attention network model; and determine the long-term encoding vector corresponding to the long-term feature group according to the similarities between the first elements in the long-term feature group.

11. The apparatus according to claim 8 , wherein the short-term feature group comprises a plurality of short-term sub-type feature groups, and the user embedding vector determining code is further configured to cause at least one of the at least one processor to determine short-term embedding sub-vectors corresponding to the plurality of short-term sub-type feature groups according to similarities between the

Assignees

• Tencent Tech Shenzhen Co Ltd

Inventors

• Qiu Zhi Jie
• Rao Jun
• Liu Yi
• Zhang Bo

Classifications

• G06F16/9535 Primary

  Search customisation based on user profiles and personalisation · CPC title