大规模动态多维社会网络的社区发现算法研究

Studying community structures has a very important theoretical significance. It helps us understand the structures and functions of networks, reveal the implicit patterns, analyze and predict the network behaviors. Meanwhile, studying community structures has a wide range of application prospects. It is able to be applied to many fields, such as business intelligence, intelligent recommendation， and social marketing. However, the characteristics of large scale, dynamic and multi-dimensional of social networks, are putting forward the new challenges to the existing methods of community detection. This project aims to study the effective and efficient algorithms to detect communities in large scale, dynamic and multi-dimensional social networks. We first study and effective data modeling methods based on tensor. Based on multi-dimensional model, we explore the evaluating metrics and study the algorithm to detect multi-dimensional communities. Based on incremental machine learning techniques, we explore some incremental algorithms to detect dynamic communities, multi-dimensional dynamic communities. Furthermore, we study the optimizing techniques to improve the complexity and parallelism of the above algorithms. With the above theoretical studies, we will design and develop a software system which is oriented to community detection in large scale and complex social networks.The work of this project is significant to the theoretical research and practical applications. It provides the theoretical foundations for the social network mining. Moreover, It is able to provide some basis for the scientific analysis of advitising, personalized recommendation, precision marketing etc. Therefore, this work is of great theoretical significance and practical application.

社区结构研究，对理解网络结构和功能、揭示网络中的隐含模式、分析及预测网络行为等具有非常重要的理论意义。同时，还可以应用在智能推荐、精准营销等诸多领域，具有广泛的应用价值。然而，网络的大规模性、动态性、多维性等特点，对现有的社区发现算法提出了挑战。本课题拟针对大规模、动态、多维社会网络，研究如下内容：(1)探索有效的多维网络建模方法；(2)基于多维网络建模，研究多维社区的评价方法及发现算法；(3)基于增量学习技术，研究增量的动态社区发现算法以及增量的多维社区动态发现算法；(4)结合算法机理和并行性分析，研究适用于大规模、动态、多维社会网络的算法优化技术。最后，基于上述研究结果，设计开发一个面向大规模复杂网络的社区发现软件系统。本课题的研究结果不仅能够为社会网络分析与挖掘领域提供理论基础，而且还能够为广告投放、个性化推荐、精准营销等相关领域提供科学分析的依据，具有重要的理论意义和应用价值。

社交媒体数据分析与挖掘是近几年来一个热门的研究领域，社区发现及分析是其中一个重要的研究课题。社区结构研究，对理解网络结构和功能、揭示网络中的隐含模式、分析及预测网络行为等具有非常重要的理论意义。同时，还可以应用在智能推荐、精准营销等诸多领域， 具有广泛的应用价值。然而，网络的大规模性、动态性、多维性等特点，对现有的社区发现算法提出了挑战。.为解决上述挑战性问题，本项目对社区的多维属性、社交个体及群体的动态性、社区的动态性进行了探索性研究，具体的研究内容包括“微博数据的管理与挖掘系统平台”、“社交媒体数据中网络流行语的发现与时空分析”、“文本分类中的基于最大化全局信息增益的特征选择方法”、“基于种子及半监督学习的重叠社区发现方法”、“面向线下信用卡数据的pos终端机的定位方法”、“社交媒体中用户地理位置模式的时空分析”、“社交媒体中流感的测与分析方法”等。.团队成员在《Knowledge-Based Systems》、《International Journal of Data Warehousing and Mining》、《International Journal of Machine Learning and Cybernetics》、《Journal of Intelligent & Fuzzy Systems》、ICDM、DASFAA、PAKDD等国内外领域重要期刊及会议上发表论文13篇，其中SCI检索5篇，EI检索9篇。.上述研究结果引起国内外同行的广泛关注，为社交网络分析与挖掘这一研究领域奠定了较好的理论基础，能够在一定程度上推动相关研究课题的发展。