摘要
近年来,学术网络经历了快速的发展.该网络结构通常包含多种类型的实体和复杂多样的实体间关系,其中作者、论文和主题是学术网络中最具代表性的三类实体,它们之间存在着类型丰富且随时间演进的交互关系.对学术网络的结构及其演进机制进行研究具有重要意义.然而,大多数相关工作仅考虑了学术网络中单一类型实体内的交互关系,如作者间的合著关系、论文间的引用关系或主题间的关联关系,未能对学术网络中不同类型的实体进行有效整合、给出对多类型实体演进学术网络进行研究分析的统一框架.为了解决上述问题,该文提出了“多实体学术模型”,将作者、论文、主题三种类型的实体整合进一个统一的理论框架,并通过研究实体间的连接关系及其演进情况来对学术网络进行刻画.其贡献点主要包括以下几个部分:(1)该文同时考虑了论文、作者、主题对学术网络性质的影响,并对包含690万条数据的微软学术网络数据集进行了统计分析,得到以下两个方面的结果:验证了一般社会网络中存在的性质如节点度服从幂律分布、幂律指数随时间收敛、网络稠密化等在多类型实体演进学术网络中同样存在;发现了一些多类型实体演进学术网络中特有的性质,如规模较大的实体往往具有更高的演进速率,幂律指数随时间的波动与收敛及实体间交互演进等,并根据其演进特性提出实体内演进、实体间演进以及交互演进三种演进模式;(2)基于上述观测现象,提出了多实体学术模型:该模型通过构建异构图的方法同时刻画同一类型实体内和不同类型实体间的连接关系,并通过直接演进、间接演进、内部演进等多种策略刻画连接关系的演进模式,具有很强的理论保证.该模型可用于多类型实体演进学术网络的数学刻画,并为学术关系预测、学术影响力传播、推荐算法设计等应用提供理论基础;(3)分别从理论分析和实验验证两个方面证明了模型的有效性:理论方面,通过数学推导证明了多实体学术模型具有节点度随时间呈多项式速率增长、节点度服从幂律分布、网络稠密化等性质;实验方面,根据多实体学术模型生成了相应的仿真网络并证明其具有上述性质.
Recent years have witnessed the rapidly growing scholarly information.All the information,when combined together,leads to the formation of the scholarly network that contains three major entities,i.e.,paper,author and topic.All the three entities interact with each other as time goes by,which results in an evolving multi-entity scholarly network.As a matter of fact,studying properties of scholarly networks and getting insight of their evolving mechanism have important implications.However,most works focus on single entity of the network,e.g.,sub-networks generated by co-authorship,citation or topic relationship;while few of them merge multiple kinds of entities into one single fabric to obtain the understanding of scholarly networks from an overall perspective.To bridge this gap,we are motivated to give the model that incorporates entities of paper,author and topic into one single framework—Multi-entity Scholarly Model(MSM),which amalgamates entities of author,paper and topic into a framework to simulate interactions among different entities,and thus presenting the co-evolution within scholarly networks.Our contributions are listed as below.(1)Our first contribution is to originally explore comprehensive properties in scholarly networks with the concern of multiple entities,i.e.,paper,author and topic.Based on scholarly datasets provided by Microsoft,which contain about 6.9 million papers,we use data-mining and other big-data analyzing approaches to observe patterns in the growth of the scholarly network.On one hand,we observe some similar features to those that have already been discovered in many traditional social networks,such as power-law degree distribution,densification,etc.On the other hand,there also exists several unique features in scholarly networks,like faster growth rate of the entity that has a bigger size,varying and converging exponents in power-law distributions with time,and the simultaneous co-evolution of all entities,etc.All these evolving features,with the awareness of multiple entities,can be categorized into three types,i.e.,inter-evolution,intra-evolution as well as the co-evolution on the whole.(2)Given empirical observations,our next significant contribution is for the first time establishing a comprehensive modeling of evolving scholarly networks.Combining entities of paper,author and topic in singe fabric,the proposed model captures both the inter-correlation and intra-correlation of the three entities during the evolving process.Particularly,inter-correlation is characterized through tripartite graph whose evolving process follows the mode of preferential attachment and intra-correlation of nodes within each entity is described as intra-degree(which we define)power-law distribution,degree densification,etc.The model can be used in characterizing evolving multi-entity scholarly networks and provides theoretical guarantee for applications such as relation prediction,influence propagation and recommendation.(3)Our third contribution is to validate the effectiveness of MSM through both theoretical analysis and empirical simulation.Based on the constructing methods of random arrival,preferential attachment,edge copying and the assumption of the affiliation relationship inside entities,we successfully obtain the growing rate of nodes’degree,power-law distributions inside or among multiple entities and the densification of the entire network.Further,we also implement simulating approaches to validate that our model can accurately reproduce real scholarly networks.
作者
刘佳琪
傅洛伊
孔令坤
甘小莺
王新兵
LIU Jia-Qi;FU Luo-Yi;KONG Ling-Kun;GAN Xiao-Ying;WANG Xin-Bing(School of Computer Science,Northwestern Polytechnical University,Xi’an 710129;School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240)
出处
《计算机学报》
EI
CSCD
北大核心
2020年第10期1791-1809,共19页
Chinese Journal of Computers
基金
国家重点研发专项(2018YFB1004705,2018YFB2100302)
国家自然科学基金(61822206,61532012,61602303,61829201,61960206002)
腾讯犀牛鸟(20180116)资助.
关键词
学术网络
演进网络
多类型实体
异构连接
交互演进
scholarly networks
evolving networks
multi-entity
heterogeneous connections
co-evolution
