Method and apparatus for generating recommendation model, content recommendation method and apparatus, device and medium

Patent metadata

Field	Value
Publication number	US-12165059-B2
Application number	US-202117171507-A
Country	US
Kind code	B2
Filing date	Feb 9, 2021
Priority date	Jun 11, 2020
Publication date	Dec 10, 2024
Grant date	Dec 10, 2024

Abstract

The present disclosure provides a method for generating a recommendation model, a content recommendation method, and a content recommendation apparatus, and an electronic device, and relates to an artificial intelligence field and a deep learning field. The method for generating a recommendation model includes: obtaining a graph training sample set; inputting the graph training sample set into a machine learning model to train the machine learning model, in which the machine learning model includes at least one low-rank graph convolutional network, and the low-rank graph convolutional network includes a complete weight matrix composed of a first low-rank matrix and a second low-rank matrix; in which a training objective of the low-rank graph convolutional network includes a first parameter item, a second parameter item and a non-convex low-rank item; and in responding to detecting that a training end condition is met, determining the machine learning model as a recommendation model.

First claim

What is claimed is:

1. A method for generating a recommendation model, comprising:
obtaining a graph training sample set comprising a first relation matrix for users, a second relation matrix for recommended contents, and a third relation matrix for the users and the recommended contents;
inputting the graph training sample set into a machine learning model to train the machine learning model, wherein the machine learning model comprises at least one low-rank graph convolutional network, and the low-rank graph convolutional network comprises a complete weight matrix composed of a first low-rank matrix and a second low-rank matrix; wherein a training objective of the low-rank graph convolutional network comprises a first parameter item for the first low-rank matrix, a second parameter item for the second low-rank matrix and a non-convex low-rank item, and the non-convex low-rank item can be decomposed into a combination of the first low-rank matrix and the second low-rank matrix; and
in responding to detecting that a training end condition is met, determining the machine learning model trained currently as a recommendation model matching the graph training sample set;
wherein the training objective of the low-rank graph convolutional network comprises:

min

Θ
r

,

Θ
c

λ
⁢



Θ
r



F
2

+

λ
⁢



Θ
c



F
2

+

f
⁡
(

W

Θ
r

⁢

H

Θ
c

T

,
Y

)

+

1
2

⁢

(



W

Θ
r



F
2

+



H

Θ
c



F
2

)

-



W

Θ
r

⁢

H

Θ
c

T



F

,

Θ r is the first parameter item, Θ c is the second parameter item, ∥⋅∥ F represents a Frobenius norm of a matrix in, λ is a hyper-parameter, f(W Θ r H Θ c T ,Y) is a loss function in the machine learning model; W Θ r is the first low-rank matrix under the first parameter item Θ r , H Θ c , is the second low-rank matrix under the second parameter item Θ c ; and W Θ r H Θ c T =X Θ , X Θ is a weight matrix under all parameter items Θ, and Y is known information input into the low-rank graph convolutional network.

2. The method according to claim 1 , wherein the non-convex low-rank item is a difference matrix between a nuclear norm of the complete weight matrix and a Frobenius norm of the complete weight matrix.

3. The method according to claim 1 , wherein the loss function f(W Θ r H Θ c T ,Y) is represented by:
f ( W Θ r H Θ c T ,Y )=∥Ω⊙( W Θ r H Θ c T −Y )∥ F 2

where Ω is a position of observation data in Y.

4. The method according to claim 1 , wherein in the process of training the machine learning model, a stochastic gradient descent method is applicable to learn the first parameter item and the second parameter item in each low-rank graph convolutional network.

5. A content recommendation method, comprising:
obtaining a first relation matrix for users and a second relation matrix for recommended contents;
inputting the first relation matrix and the second relation matrix into the recommendation model pre-trained by the method of claim 1 , to obtain a third relation matrix for the users and the recommended contents output by the recommendation model;
obtaining a recommended content with a highest weight corresponding to a user from the third relation matrix as a target recommended content; and
recommending the target recommended content to the user.

6. An electronic device, comprising:
at least one processor; and
a memory connected in communication with the at least one processor; wherein,
the memory stores instructions executable by the at least one processor, when the instructions are executed by the at least one processor, the at least one processor is caused to implement the method for generating a recommendation model or the content recommendation method;
wherein the method for generating a recommendation model comprises: obtaining a graph training sample set comprising a first relation matrix for users, a second relation matrix for recommended contents, and a third relation matrix for the users and the recommended contents; inputting the graph training sample set into a machine learning model to train the machine learning model, wherein the machine learning model co

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Wang Yaqing
• Xiong Hui

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0495

  Quantised networks; Sparse networks; Compressed networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06F18/2133

  based on naturality criteria, e.g. with non-negative factorisation or negative correlation · CPC title