Content processing method and apparatus, computer-readable storage medium and computer device

Patent metadata

Field	Value
Publication number	US-12001494-B2
Application number	US-202217692368-A
Country	US
Kind code	B2
Filing date	Mar 11, 2022
Priority date	Feb 13, 2020
Publication date	Jun 4, 2024
Grant date	Jun 4, 2024

Abstract

Embodiments of the present disclosure relate to a content processing method and apparatus, a computer-readable storage medium, and a computer device. The method includes obtaining data of a to-be-pushed user in behavior evaluation dimensions, user portrait information of the to-be-pushed user, and to-be-pushed content, with the user portrait information corresponding to a user portrait code and the to-be-pushed content corresponding to a to-be-pushed content code. The method further includes determining a user behavior code based on the data of the to-be-pushed user in the behavior evaluation dimensions, and obtaining a target code by fusing the to-be-pushed content code with the user behavior code and the user portrait code. Then, the method includes determining a recommendation probability of the to-be-pushed content based on the target code to determine the push content based on the recommendation probability.

First claim

What is claimed is:

1. A content processing method, performed by at least one processor, the method comprising:
obtaining data of a to-be-pushed user in behavior evaluation dimensions, user portrait information of the to-be-pushed user, and to-be-pushed content, wherein the user portrait information matches a corresponding user portrait code, and the to-be-pushed content matches a corresponding to-be-pushed content code;
determining a user behavior code of the to-be-pushed user in the behavior evaluation dimensions based on the data of the to-be-pushed user in the behavior evaluation dimensions;
obtaining a target code based on the to-be-pushed content code, the user behavior code, and the user portrait code;
determining a recommendation probability of the to-be-pushed content based on the target code; and
determining push content corresponding to the to-be-pushed user from the to-be-pushed content based on the recommendation probability,
wherein the behavior evaluation dimensions comprise a first behavior evaluation dimension and a second behavior evaluation dimension,
wherein the first behavior evaluation dimension uses a first relationship between the to-be-pushed user and the push content in a preset period of time, and
wherein the second behavior evaluation dimension uses a second relationship between the to-be-pushed user and a publishing object of the push content in the preset period of time.

2. The method of claim 1 , wherein the obtaining data of the to-be-pushed user in the behavior evaluation dimensions comprises:
acquiring operation data of the to-be-pushed user on the push content in the preset period of time, association data between the to-be-pushed user and the publishing object of the push content in the preset period of time, and operation data of the publishing object on the push content in the preset period of time; and
obtaining data of the to-be-pushed user in the behavior evaluation dimensions in the preset period of time based on the operation data of the to-be-pushed user on the push content in the preset period of time, the association data between the to-be-pushed user and the publishing object of the push content in the preset period of time, and the operation data of the publishing object on the push content in the preset period of time.

3. The method of claim 2 , wherein the behavior evaluation dimensions further comprises a third behavior evaluation dimension; and
wherein the obtaining data of the to-be-pushed user in the behavior evaluation dimensions in the preset period of time comprises:
obtaining the first relationship between the to-be-pushed user and the push content based on the operation data of the to-be-pushed user on the push content in the preset period of time, and using the first relationship between the to-be-pushed user and the push content as data of the to-be-pushed user in the first behavior evaluation dimension;
obtaining the second relationship between the to-be-pushed user and the publishing object based on the association data between the to-be-pushed user and the publishing object of the push content in the preset period of time, and using the second relationship between the to-be-pushed user and the publishing object as data of the to-be-pushed user in the second behavior evaluation dimension; and
obtaining a third relationship between the publishing object and the push content based on the operation data of the publishing object on the push content in the preset period of time, and using the third relationship between the publishing object and the push content as data of the to-be-pushed user in the third behavior evaluation dimension.

4. The method of claim 1 , wherein the data of the to-be-pushed user in the behavior evaluation dimensions comprises data objects and data relationships between the data objects; and
the determining the user behavior code of the to-be-pushed user in the behavior evaluation dimensions based on the data of the to-be-pushed user in the behavior evaluation dimensions comprises:
constructing a behavior graph network of the to-be-pushed user by using the data objects as nodes and the data relationships between the data objects as edges of the nodes; and
determining the user behavior code of the to-be-pushed user in the behavior evaluation dimensions based on the behavior graph network of the to-be-pushed user.

5. The method of claim 4 , wherein the nodes in the behavior graph network match corresponding initial information codes; and
the determining the user behavior code of the to-be-pushed user in the behavior evaluation dimensions based on the behavior graph network of the to-be-pushed user comprises:
inputting the initial information codes of the nodes in the behavior graph network of the to-be-pushed user into a pre-trained information embedding network model to obtain the user behavior code of the to-be-pushed user in the behavior evaluation dimensions, wherein the pre-trained information embedding network model is configured to obtain the user behavior code of the to-be-pushed user in the behavior evaluation dimensions based on aggregating target information codes of neighbor nodes of the node in which the to-be-pushed user is located, and concatenating a target information code of the neighbor node of the node in which the to-be-pushed user is located obtained after the aggregation and the initial information code of the node in which the to-be-pushed user is located, and
wherein the target information code of the neighbor node of the node in which the to-be-pushed user is located being obtained by aggregating target information codes of neighbor nodes of the neighbor node and concatenating a target information code of the neighbor node of the node in which the to-be-pushed user is located obtained after the aggregation and the initial information code of the neighbor node of the node in which the to-be-pushed user is located.

6. The method of claim 5 , wherein the determining the recommendation probability of the to-be-pushed content based on the target code comprises:
inputting the target code into a pre-trained content recommendation model, and performing convolution pooling processing on the target code by using the pre-trained content recommendation model, to obtain the recommendation probability of the to-be-pushed content.

7. The method of claim 6 , wherein training the pre-trained information embedding network model and the pre-trained content recommendation model comprises:
obtaining data of a sample user in the behavior evaluation dimensions, sample user portrait information, sample push content, and an actual recommendation probability of the sample push content;
respectively inputting the sample user portrait information and the sample push content into a feature extraction network in a to-be-trained content recommendation model, to obtain a sample user portrait code corresponding to the sample user portrait information and a sample push content code corresponding to the sample push content;
inputting the data of the sample user in the behavior evaluation dimensions into a to-be-trained information embedding network model, to obtain a sample user behavior code of the sample user in the behavior evaluation dimensions;
fusing the sample push content code with the sample user behavior code and the sample user portrait code to obtain a sample target code;
inputting the sample target code into a content prediction network in the to-be-trained content recommendation model to obtain a predicted recommendation probability of the sample push content;
calculating a loss value based on the predicted recommendation probability and the actual recommendation probability;
adjusting network parameters of the to-be-trained information embedding network model an

Assignees

• Tencent Tech Shenzhen Company Ltd

Inventors

• Du Lihuan
• Wang yalong
• Wang Rui
• Xia Feng
• Lin Leyu

Classifications

• G06F16/9535 Primary

  Search customisation based on user profiles and personalisation · CPC title

• G06F16/335 Primary

  Filtering based on additional data, e.g. user or group profiles (filtering in web context G06F16/9535, G06F16/9536) · CPC title