CAREER: Novel Data Mining Technologies for Complex Network Analysis

长期的研究目标是开发新的数据挖掘技术，以阐明复杂但无处不在的网络的结构和动力学。复杂网络是由非平凡关系(边)连接的元素(顶点)的大系统。这种复杂网络的例子包括WWW、新陈代谢和蛋白质网络、社会网络以及经济和金融市场。网络系统的基本原理和规律可以帮助我们构建更有效的通信机制，找到致命疾病的治疗方法，并处理经济危机。尽管在理解支配复杂网络结构和行为的基本规律方面取得了重大进展，但当前的分析技术与它们对现实世界复杂网络的适用性之间仍然存在脱节。缺乏一种原则性的方法来系统地分析单个大型复杂网络，并将系统行为与网络结构联系起来。还迫切需要一个理论框架来理解多个网络之间的关系，这是比较网络分析的关键。如何集成和利用与网络拓扑相关的丰富的系统数据，如测量时间序列，来研究复杂系统仍然是一个悬而未决的问题。该项目通过以下方式解决这些问题：1)开发新的图形和信息理论方法来提取既简化和突出网络结构的网络主干；2)开发用于比较网络分析的信息理论网络距离度量和聚类算法；3)应用因果推理和网络模块将时间序列与网络拓扑相结合。所提出的挖掘方法建立在图论、信息论和统计学习概念和技术的创新融合之上，可以极大地扩展数据挖掘的范围。这个项目还将帮助我们更好地分析新出现的复杂、异质和大规模的现实世界复杂网络数据。在社会和政治科学、软件工程和生物信息学领域的专家的密切合作下，所提出的技术有可能帮助理解人类社会是如何在个人层面(社会网络)和组织层面(政治学)组织起来的；阐明大型软件系统是如何形成和演化的；揭示动态环境中生物细胞系统的组织原理；以及识别治疗或药物靶标。利用流行的在线社交网络，如MySpace和Facebook，将吸引、招募和培养来自代表不足的群体的学生，包括妇女和少数族裔，并让他们参与尖端研究。有关更多信息，请参阅项目网页：http://www.cs.kent.edu/~jin/NSFCAREER/

The long-term research goal is to develop novel data miningtechnologies to elucidate the structures and dynamics of complexbut ubiquitous networks. A complex network is a large systemof elements (vertices) that are joined by non-trivial relationships(edges). Examples of such complex networks include the WWW,metabolic and protein networks, social networks, and economicand financial markets. The underlying principles and laws of thesenetwork systems can help us construct more effectivecommunication mechanisms, find cures for fatal diseases, anddeal with economic crises.In spite of the significant advances that have been made towardsunderstanding the fundamental laws that govern the structure andbehavior of complex networks, there is still a disconnect between current analytical techniques and their applicability to real-worldcomplex networks. A principled approach is lacking to systematicallyanalyze a single large complex network and link system behaviors to network structure. There is also immediate and crucial need for atheoretical framework to understand the relationships betweenmultiple networks, which is the key for comparative network analysis. How to integrate and leverage the rich system data, such asmeasurement time series, associated with network topology to studycomplex systems is still an open question. This project addressesthese issues by: 1) developing novel graph and informationtheoretical approaches to extracting network backbones which bothsimplify and highlight network structures; 2) developing informationtheoretical network distance measures and clustering algorithms forcomparative network analysis; 3) applying causality inference andnetwork modularity to integrate time series with network topology.The proposed mining methodologies build upon an innovative blendof graph theoretical, information theoretical, and statistical learningconcepts and techniques, and can greatly expand the reach ofdata mining. This project will also help us better analyze theemerging complexity, heterogeneity, and large scale of real-worldcomplex network data.In a close collaboration with domain experts from social and politicalsciences, software engineering, and bioinformatics, the proposedtechniques have the potential to help understand how human society isorganized at the individual level (social networks) and organizationallevel (political science); illuminate how large scale software systemsform and evolve; reveal the organizational principles of biocellularsystems in a dynamic environment; and identify therapeutic or drugtargets. Using the popular online social networks, such as MySpaceand Facebook, as ``hooks'', this project will attract, recruit, andprepare students from underrepresented groups including womenand minorities to computer science and involve underrepresentedstudents in the cutting-edge research.For further information see the project web page at:http://www.cs.kent.edu/~jin/NSFCAREER/