半监督全局异构信息保存的网络表示学习  

NETWORK REPRESENTATION LEARNING BASED ON SEMI SUPERVISED GLOBAL HETEROGENEOUS INFORMATION PRESERVATION

作　　者：李征[1,2] Li Zheng(School of Electronic Engineering,Changzhou Institute of Information Technology,Changzhou 213164,Jiangsu,China;School of Electronic Information,Jiangsu University of Science and Technology,Zhenjiang 212003,Jiangsu,China)

机构地区：[1]常州信息职业技术学院电子工程学院,江苏常州213164 [2]江苏科技大学电子信息学院,江苏镇江212003

出　　处：《计算机应用与软件》2023年第6期257-265,共9页Computer Applications and Software

基　　金：2019年江苏省“青蓝工程”优秀教学团队(计算机应用技术)项目(GB20190342);2017年江苏省职业院校高水平骨干专业(计算机应用技术)建设项目(SJG201717131);2019年度江苏省常州市常州信息职业技术学院自然科学重点科研项目(CXKZ201909Z);2019年度江苏省常州市常州信息职业技术学院青年自然基金项目(CXKZ201910Q)。

摘　　要：传统的网络表示学习模型存在网络信息保存不全面的缺点,为此提出一种半监督全局异构信息保存网络表示学习框架。采用元路径引导的随机行走策略对路径进行采样并提取异构信息,从而通过自定义的路径模式形式来挖掘异构信息网络中的隐藏信息。该方法通过最小化自编码器的节点检索误差来获得预边缘节点表示,推断节点的边缘属性,并在一个连续的公共向量空间中表示不同类型的节点。进一步通过灵活的自动编码网络,将节点间的高阶邻近性和语义信息集成到一个通用框架中。通过几个异构信息网络挖掘任务,实验结果表明,该方法能够有效地保存各种路径模式下的网络信息,在网络重构、链路预测和节点分类等实际应用中表现优于现有的方法。The traditional network representation learning model has the disadvantage of incomplete network information preservation.Therefore,a semi supervised global heterogeneous information preservation network representation learning framework is proposed.The random walk strategy guided by meta path was used to sample the path and extract the heterogeneous information,so as to mine the hidden information in heterogeneous information network through the self-defined path pattern.A method was proposed to obtain the pre-edge node representation by minimizing the node retrieval error of the self-encoder,to infer edge attributes of nodes,and to represent different types of nodes in a continuous common vector space.Furthermore,the high-order proximity and semantic information between nodes were integrated into a general framework through a flexible automatic coding network.Through several heterogeneous information network mining tasks,the experiments show that this method can effectively save the network information under various path patterns,and is better than the existing methods in network reconstruction,link prediction,node classification and other practical applications.

关 键 词：半监督 异构信息 网络表示学习 信息保存 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]