基于集合约束的异质超网络表示学习  

Heterogeneous Hypernetwork Representation Learning Based on Set Constraints

作　　者：刘贞国 朱宇 王晓英[1] 黄建强[1] 曹腾飞[1] LIU Zhenguo;ZHU Yu;WANG Xiaoying;HUANG Jianqiang;CAO Tengfei(Department of Computer Technology and Application,Qinghai University,Xining 810000,China)

机构地区：[1]青海大学计算机技术与应用系,西宁810000

出　　处：《计算机工程》2023年第11期115-122,共8页Computer Engineering

基　　金：国家自然科学基金(62166032,62162053,62062059,62101299);青海省自然科学基金(2022-ZJ-961Q)。

摘　　要：与节点之间仅具有成对关系的普通网络不同,超网络的节点之间还存在复杂的元组关系,因而现有的大多数普通网络表示学习方法不能有效地捕获复杂的元组关系。为此,提出一种捕获成对关系和元组关系的基于集合约束的异质超网络表示学习方法。结合团扩展和星型扩展,将抽象为超图的异质超网络转化成抽象为2-截图+关联图的异质网络。基于2-截图+关联图,采用感知节点语义相关性的元路径游走方法获取异质节点序列,并通过基于拓扑派生目标函数的模型训练异质节点序列上的成对关系,采用基于集合约束目标函数的模型,将与节点关联的超边集合融入到超网络表示学习中来训练节点之间的元组关系,从而获得高质量的节点表示向量。实验结果表明,对于链接预测任务,该方法的性能接近于其他最优基线方法;对于超网络重建任务,当超边重建比率大于0.7时,该方法在drug数据集上具有较优的性能,在GPS数据集上的平均性能超过其他最优基线方法16.2%。ed as a 2-section graph+incidence graph.Based on this combination of 2-section graph+incidence graph,the meta-path walk method is employed to consider the semantic relevance of nodes,generating heterogeneous node sequences.Pairwise relationships within these sequences are then trained using a model grounded in topology-derived objective functions.Finally,the model,including a set constraint objective function,incorporates hyperedge sets associated with nodes into the hypernetwork representation learning process to train tuple relationships among nodes,thereby yielding high-quality node representation vectors.Experimental results demonstrate that,for link prediction tasks,the proposed method performs comparably to other optimal baseline methods.In hypernetwork reconstruction tasks,the method outperforms other optimal methods on the drug dataset when the hyperedge reconstruction ratio exceeds 0.7.Additionally,the average performance of the proposed method surpasses other optimal baseline methods by 16.2%on the GPS dataset.

关 键 词：网络表示 超网络结构 集合约束 链接预测 超网络重建 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]