Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

News recommendation method and system based on graph neural network

A neural network and recommendation method technology, applied in the field of news recommendation, can solve the problems of not being able to meet the timeliness of news recommendation, lack of effectively capturing user behavior preferences and combining content preferences, etc., to achieve the effect of accurate, effective, and time-sensitive recommendation.

Pending Publication Date: 2020-07-03
UNIV OF ELECTRONIC SCI & TECH OF CHINA
View PDF5 Cites 12 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problems that the existing technology cannot satisfy the timeliness of news recommendation and the lack of effective capture of user behavior preference and combination of content preference when recommending, the present invention proposes a news recommendation method and system based on graph neural network. The network combines the probability matrix decomposition to model the user's historical click behavior information, realizes the rapid news recommendation and effectively combines the user's behavior preferences and content preferences to make recommendations

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • News recommendation method and system based on graph neural network
  • News recommendation method and system based on graph neural network
  • News recommendation method and system based on graph neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0029] The present invention proposes a news recommendation method based on graph neural network, such as figure 1 As shown, firstly, the user-news interaction structure diagram and user-news behavior interaction matrix are respectively generated through the user behavior sequence, and then the global target news feature vector and the local target news feature vector are generated from the user-news interaction behavior structure chart and the user-news behavior interaction matrix News feature vectors, user feature vectors, and then linearly superimpose the global target news feature vectors and local target news feature vectors into target news feature vectors, and finally perform inner product on target news feature vectors and user feature vectors to obtain users’ prediction scores for news; For each user, the news score prediction will be performed, and the predicted scores of all news will be sorted, and the top K ones will be selected for recommendation.

Embodiment 2

[0031] The present invention is on the basis of above-mentioned embodiment 1, as figure 2 As shown, further, the specific construction method of the constructed user-news interaction behavior structure diagram is as follows: first, assume that the hypothetical system generates the event sequence e 1 ,e 2 ,e 3 ...e n , and each event contains the user's session information, that is, user id, session start flag, session stop flag, and occurrence time; we first define a news item as a node in the user-news interaction behavior structure graph. Then, according to the daily log sequence, a unique user click record is constructed for each user, with the user id as the primary key. We scan each record from front to back, and use "sessionstart flag" and "session stop flag" as segmentation marks to divide the user sequence into different user behavior segments. From each behavior segment, a binary sequence pair is generated as an edge of the user-news interaction behavior structur...

Embodiment 3

[0034] The present invention is on the basis of any one of above-mentioned embodiment 1-2, in order to realize the present invention better, as image 3 As shown, further, the steps of generating the global target news feature vector are as follows: first construct the user-news interaction behavior structure diagram; Node2Vec) to get the global target news feature vector. The node vectorization vector model (Node2Vec) includes three steps: the depth walk algorithm generates the node sequence, the vertex sampling algorithm (Alias) performs vertex sampling, and the node sequence uses the word vector extraction algorithm to generate the vector encoding of the vertex.

[0035] Working principle: The specific operation of constructing the user-news interaction behavior structure diagram is as described in claim 2, so it will not be repeated; after obtaining the user-news interaction behavior structure diagram, the node sequence is generated through the depth walk algorithm, and the ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention provides a news recommendation method and system based on a graph neural network. Firstly, a user-news interaction behavior structure chart is generated through a user behavior sequenceof a user behavior generation layer; a user-news behavior interaction matrix is generated through a behavior matrix generation layer, then a user-news interaction behavior structure chart is processedby a node vectorization model layer to generate a global target news feature vector, and the user-news behavior interaction matrix is processed by a probability matrix decomposition layer to obtain alocal target news feature vector and a user feature vector; linearly superposing the global target news feature vector and the local target news feature vector into a target news feature vector by afusion layer, and performing inner product on the target news feature vector and the user feature vector to obtain a predicted score of the news by the user; and sorting the prediction scores of all news, and selecting the first K news in sequence to recommend. According to the method, timeliness of news recommendation is met, and behavior preferences and content preferences of users are effectively combined.

Description

technical field [0001] The invention belongs to the field of news recommendation, and in particular relates to a graph neural network-based news recommendation method and system thereof. Background technique [0002] Recommender systems have been widely used in web-based news applications, such as Toutiao, Tencent News, Netease News, Sohu News, etc., to handle information overhead by analyzing rich user information and news text information and other heterogeneous content. Traditional news recommendation methods mostly focus on analyzing news content as the main entry point, and use topic models and more complex deep learning models to learn vectorized representations of news, such as item-based collaborative filtering algorithm (CB), recommendation algorithm based on matrix decomposition ( MF), collaborative topic regression algorithm (CTR) and collaborative variational autoencoder algorithm (CVAE). [0003] Due to the increasing difficulty of obtaining user data and the i...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06F16/9535G06N3/04G06N3/08
CPCG06F16/9535G06N3/08G06N3/045
Inventor 李赵宁钱伟中杨茂林
Owner UNIV OF ELECTRONIC SCI & TECH OF CHINA
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com