异构信息网络中节点相似性搜索并行算法研究  

Research on Parallel Algorithm for Node Similarity Search in Heterogeneous Information Network

作　　者：徐小玉 陈仲委[1] XU Xiaoyu;CHEN Zhongwei(Zhejiang Wanli University,Ningbo Zhejiang 315100)

机构地区：[1]浙江万里学院,浙江宁波315100

出　　处：《浙江万里学院学报》2024年第3期82-90,共9页Journal of Zhejiang Wanli University

基　　金：浙江省社会科学界联合会研究课题成果“共同富裕视域下社区精准志愿服务体系的研究与实现”(2023N075)。

摘　　要：传统的基于串行计算的节点相似性搜索算法,在处理增量式异构信息网络数据流时面临效率低下、资源消耗过大等问题。为此,文章利用同构信息网络中并行约简的基本原理与基本方法,定义了异构信息网络的并行约简度量方法:值相似依赖度和余弦实体相似依赖度,提出异构数据流中元路径下节点相似性搜索并行算法(FPathSim),F-PathSim可以并行约简、并行计算,整体上删除对节点相似性搜索冗余的数据,减少对整个数据集的重复处理。在DBLP数据集上进行大量的实验,实验结果表明F-PathSim能较好的适应异构信息网络中增量式数据流中节点相似性搜索要求。The traditional node similarity search algorithm based on serial computing has issues with low efficiency and excessive resource consumption,especially when dealing with incremental heterogeneous information network data flow.Therefore,this paper utilizes the basic principles and methods of parallel reduction in isomorphic information networks to define the parallel reduction measurement methods of heterogeneous information networks:value similarity dependency and cosine entity similarity dependency.It proposes a parallel algorithm for node similarity search under meta-paths in heterogeneous data streams(F-PathSim).F-PathSim can perform parallel reduction and parallel computation,deleting redundant data for node similarity search on the whole and reducing repeated processing of the entire data set.A large number of experiments were conducted on the DBLP data set,and the experimental results showed that FPathSim can better meet the requirements of node similarity search in incremental data streams in heterogeneous information networks.

关 键 词：异构信息网络 同构信息网络 并行约简 元路径 增量式数据流 相似依赖度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]