首页|基于邻域抽样多图神经网络的社会化推荐算法

基于邻域抽样多图神经网络的社会化推荐算法

扫码查看
基于图神经网络的社会化推荐算法可以从图网络中获取深层数据信息,提升推荐性能.但随着图网络复杂度提升,特别是对于多图神经网络,节点特征获取质量直接影响最终的推荐质量.为了提升多图网络中的节点特征获取质量,结合邻域抽样思想,提出一种邻域抽样多图神经网络社会化推荐模型MGNN-NS.基于用户-项目评分图和用户社交关系图,从用户和商品项目角度对图中节点的邻域节点进行抽样,并应用多头注意力机制对抽样节点进行信息聚合,获取用户和商品项目特征,计算预测评分,得到推荐结果.在真实数据集Epinions和Ciao上进行实验,结果表明MGNN-NS模型相较于基准算法有更好的推荐效果.
Social Recommendation Algorithm Based on Multi-Graph Neural Network with Neighborhood Sampling
Social recommendation algorithms based on graph neural networks can obtain deep data information from graph networks to improve recommendation performance.However,with the increasing complexity of graph net-works,especially for the multi-graph neural network,node feature information learning directly affects the final recom-mendation quality.In order to improve the quality of node feature acquisition in multi-graph neural networks,a social recommendation model based on a multi-graph neural network with neighborhood sampling(MGNN-NS)is proposed.From the perspective of users and items,the proposed model samples and aggregates the neighborhood nodes of those nodes in the user-item rating graph and user social relation graph,and learns the characteristics of target nodes.This model also applies multi-head attention networks to reduce the influence of the mean aggregator,and finally obtains the characteristic representation of users and items to calculate the prediction scores for recommendation.Experiment results on the real-world datasets of Epinions and Ciao show that this model has better recommendation effect than the benchmark algorithms.

Graph neural networkSocial recommendationNeighborhood samplingMulti-head attention mecha-nism

王若辰、原欣伟、段刚龙、李建勋

展开 >

西安理工大学经济与管理学院,陕西 西安 710054

图神经网络 社会化推荐 邻域抽样 多头注意力机制

国家自然科学基金陕西省教育厅科研项目

7187214918JK0542

2024

计算机仿真
中国航天科工集团公司第十七研究所

计算机仿真

CSTPCD
影响因子:0.518
ISSN:1006-9348
年,卷(期):2024.41(3)
  • 25