属性网络中结合用户偏好的社区搜索和离群点检测  

Incorporating User Preference for Community Search and Outlier Detection in Attributed Network

作　　者：李青青 马慧芳 李举 李志欣[3] 姜彦斌 LI Qing-qing;MA Hui-fang;LI Ju;LI Zhi-xin;JIANG Yan-bin(College of Computer Science and Engineering,Northwest Normal University,Lanzhou,Gansu 730070,China;Guangxi Key Laboratory of Trusted Software,Guilin University of Electronic Science and Technology,Guilin,Guangxi 541004,China;Guangxi Key Lab of Multi-Source Information Mining&Security,Guangxi Normal University,Guilin,Guangxi 541004,China)

机构地区：[1]西北师范大学计算机科学与工程学院,甘肃兰州730070 [2]桂林电子科技大学广西可信软件重点实验室,广西桂林541004 [3]广西师范大学广西多源信息挖掘与安全重点实验室,广西桂林541004

出　　处：《电子学报》2022年第9期2172-2180,共9页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61762078,No.61363058,No.6196604);广西多源信息挖掘与安全重点实验室开放基金(No.MIMS18-08);西北师范大学2019年度青年教师科研能力提升计划重大项目(No.NWNU-LKQN2019-2)。

摘　　要：社区搜索是备受关注的网络分析任务之一,旨在搜寻包含查询节点的局部社区.现有大多数社区搜索方法多面向简单网络且仅能定位查询节点所在社区,未能在搜索过程中考虑用户偏好.为实现利用用户偏好指导搜索过程并搜寻用户感兴趣的多社区,设计了属性网络中结合用户偏好的社区搜索和离群点检测方法,旨在通过较少的查询节点有效的捕获用户偏好并自动探索网络中的社区,同时识别社区中离群点.具体而言,通过编码查询节点及其邻居间的显式交互关系和相似属性以突出局部结构,利用其来挖掘潜在查询节点候选集成员.在查询节点候选集上定义平均划分相似度以推断属性子空间为用户潜在兴趣.采用属性和结构约束来搜索网络中的多社区和离群点.此外,真实数据集和人工数据集上的大量实验证明了所提方法的有效性.Community search aims to search local communities containing query nodes,which is one of the most concerned studies in network analysis task.Most existing community search methods are oriented to simple network and can only detect the community where query nodes are located.They may fail to take user's preferences into account during searching process.To guide the process of community search via user's preferences for finding multi-communities that us⁃ers are interested in,we propose a community search method that is capable of searching multi-communities with user's preference and simultaneously identify outliers via few given query nodes in attributed network.Clearly,we explicitly mod⁃el interactions between query nodes with its neighbors and encode similar attributes to highlight the local structure,which could be beneficial for query nodes to mine potential candidates.And we define the average partition similarity on candi⁃date set of query nodes to infer attribute subspace as user's latent interest.Multi-communities and outliers in the whole net⁃work are detected via fractional-core and structural constraints.Experiments on real and synthetic network datasets demon⁃strate the effectiveness of the proposed algorithm.

关 键 词：属性网络 社区搜索 平均划分相似度 属性子空间 离群点 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP274[自动化与计算机技术—控制科学与工程]