Graph model training, method and device for graph model based recommendation

By constructing a ternary heterogeneous graph model of users-resources-community, the problem of existing recommendation systems' difficulty in capturing cross-domain interests and preferences in user-generated content ecosystems is solved, achieving more accurate personalized and community-based content recommendations, and improving user stickiness and the diversity of the content ecosystem.

CN122197962APending Publication Date: 2026-06-12BEIJING BAIDU NETCOM SCI & TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
BEIJING BAIDU NETCOM SCI & TECH CO LTD
Filing Date
2026-03-03
Publication Date
2026-06-12

AI Technical Summary

Technical Problem

Existing recommendation systems rely solely on users' historical behavior data, making it difficult to effectively capture users' potential interests and preferences in cross-domain contexts. In particular, in the user-generated content ecosystem, the interaction between users and the community is not fully utilized, resulting in recommendation results that lack context awareness and fail to meet the needs of community-based content consumption.

Method used

Construct a graph information network to generate graph model training data, including user, resource, and community nodes. Train the graph model through sampling and loss function, learn node vectors, establish a user-resource-community ternary heterogeneous graph model, and combine graph neural network for representation learning to characterize user interests and group preferences in different community scenarios.

Benefits of technology

It improves the recommendation system's ability to characterize user community interests and group preferences, enhances the recommendation effect on community-based content, alleviates the problem of insufficient exposure of long-tail resources, and realizes the comprehensive recommendation of personalized and community-based content.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122197962A_ABST
    Figure CN122197962A_ABST
Patent Text Reader

Abstract

The disclosure provides a graph model training method and device, and a graph model recommendation method and device, and relates to the technical field of computers, in particular to the technical field of graph models, graph information networks, recommendation systems and the like. The specific implementation scheme is as follows: training data of a graph model is generated according to a graph information network, the training data including node samples and edge samples; the node samples include information of various types of nodes in the graph information network, and the edge samples include association relationships between different nodes in the graph information network; the training data is sampled to obtain a batch of training samples; wherein one training sample includes a target node and positive and negative samples thereof; the batch of training samples is input into the graph model to obtain node vectors of the positive and negative samples of the target node and a prediction result about the target node and the positive and negative samples; a loss function obtained according to the prediction result is used to train the graph model to output node vectors of the graph information network; wherein the loss function is constructed according to the node vectors of the positive and negative samples and the prediction result.
Need to check novelty before this filing date? Find Prior Art