针对有向网络的链路预测原理及方法研究

Link prediction aims at estimating the 1ikelihood of the existence of links between nodes in complex network．Recently, it has made great progress in the link prediction of undirected unweighted network. However, a lot of directed complex networks in nature and society still have not been fully studied. Compared with undirected network, the particularity and complexity of link prediction in directed network are described in three-fold:1)it need to estimate both likelihood of existence and direction of links between nodes;2)many important network structures mentioned by undirected similarity indices, such as common neighbor, have undergone tremendous changes in structure and function;3) external information, such as node attribute ,is giving a new meaning for direction prediction. To promote the formation of the theoretical algorithm framework for link prediction in directed networks, we focus on the following three aspects. Firstly, we transform and expand the classical undirected similarity indices. Based on the connection establishment mechanism of directed networks, some similarity indices suitable for the characteristics of directed networks will be proposed. Secondly, taking into account both network structure and external information, some new link prediction algorithms for directed network will be proposed by quantifying the influences of external information on direction and community. Thirdly, in addition to the study of similarity methods, based on game theory and other related theories an explorative study on the theory of link prediction in directed network will be provided.

链路预测旨在利用已知信息预测复杂网络中任意节点间存在连接的可能性。现有研究已在无向无权网络中取得很大进展。然而，面对自然界和人类社会中普遍存在的有向网络，其研究较少。与无向网络相比，有向网络链路预测具有一些复杂性和特殊性：1)需要同时预测连接概率和方向；2）许多方法中涉及的重要网络结构如共同邻居，都已经在结构和功能上发生了很大变化；3）外部信息如个体属性，对有向网络连边方向的预测也赋予了新的价值。为了推动有向网络的链路预测理论算法框架的形成，课题将首先在研究无向相似性指标的有向化及拓展的基础上，从网络演化的角度挖掘有向连边机理，并基于有向连边机理提出适合有向网络自身特点的相似性算法；然后，通过研究量化外部信息对链路预测方向性和社区性的约束和影响，提出结合外部信息的有向链路预测算法；最后，在相似性方法研究之外，探索性研究基于博弈论相关理论的有向网络链路预测理论方法。

链路预测旨在处理一个最基本的科学问题即：缺失信息的预测和还原，现有研究已在无向无权网络中取得很大进展。然而，面对自然界和人类社会中普遍存在的有向网络，其需要同时预测连接概率和方向，亟需开展体系化的有向网络链路预测研究。为了推动有向网络的链路预测理论算法框架的形成，课题从以下几个方面开展研究：.（1）无向相似性指标的有向化转化及拓展研究，针对有向链路预测的连边方向和概率预测，对无向指标进行转化拓展；.（2）相关有向演化机理的链路预测算法研究，提出基于模体演化的有向链路预测，并从微观、介观、宏观、时间等多视角研究链路预测理论与方法；.（3）基于信息论、影响力的链路预测原理方法研究，从信息论、影响力的角度系统性、多角度研究有向网络链路预测框架与算法；.（4）基于经济学相关理论的有向网络链路预测方法研究，基于博弈论、投资收益等经济学原理，探索有向网络链路预测研究新思路；.（5）基于链路预测的网络异常行为检测应用探索，基于链路预测方法，开展网络拓扑中随机异常行为、方向协同异常行为、时空协同异常行为等异常行为检测研究，并在实际中开展应用。.经过课题研究，提出了系统化的基于网络模体的有向网络链路预测原理及体系、基于多尺度视角的有向网络链路预测原理及体系，丰富了有向网络的演化机理及应用；基于信息论和影响力拓展链路预测的有向连边机理，以博弈论等经济学原理探索链路预测原理及方法，拓展了有向链路预测的理论边界和研究思路；针对网络拓扑异常行为检测开展链路预测应用，利用链路预测挖掘拓扑的异常，在实际中产生巨大作用，为挖掘网络空间中诈骗、欺诈等进行了实际应用，丰富了链路预测的实际应用场景。课题相关成果发表论文21篇，其中SCI检索7篇，EI检索6篇，专利撰写2篇，支撑形成河南省科技进步一等奖1项，培养战略支援部队学科拔尖人才培养对象1名、在读博士后1名，研究生20余人。