Link prediction for online social networks via graph attention network
刘渊 1杨凯 1苏嘉良 1袁铭 1赵紫娟2
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 扬州大学信息工程学院,江苏扬州 225127
2. 上海理工大学管理学院,上海 200093
折叠
摘要
随在线社交网络规模的不断增长,传统的链路预测方法难以捕获每个用户的全面特征信息.针对该问题,提出一种基于图注意力网络的链路预测方法(link prediction based on graph attention network,LP-GAT).首先,将在线社交网络表示为图结构数据,反映完整的用户属性信息和社交网络结构信息;其次,在图神经网络模型中引入注意力机制,更准确地刻画用户的社交特征信息,并生成用户节点的嵌入表示;最后,将节点嵌入表示输入分类器进行模型性能评估.在4个真实的在线社交网络数据集进行链路预测实验,结果表明所提模型较传统链路预测方法的性能更优.
Abstract
With the increasing scale of online social networks,traditional link prediction methods are difficult to capture comprehensive feature information of each user.To address this problem,a no-vel link prediction method for online social networks based on graph attention network(LP-GAT)is proposed in this paper.First,the online social network is represented as a graph to preserve the complete user attribute information and social network structural information.Secondly,the atten-tion mechanism is introduced into the graph neural network model to more accurately depict the user's social feature information and generate the embedded representation of the user node.Final-ly,the performance of the model is evaluated by feeding the node embedding representations into a classifier.The results of link prediction experiments on 4 real online social network data sets show that the proposed model has better performance than traditional link prediction methods.
关键词
在线社交网络/图神经网络/图注意力网络/链路预测
Key words
online social network/graph neural networks/graph attention network/link prediction
维普
万方数据