多维海量社交网络数据可视化技术研究

发布时间：2018-03-01 22:32

本文关键词： 社交网络数据可视化社区发现力导引布局属性图聚类　出处：《重庆邮电大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着信息科技的飞速发展,具有多维和海量等特性的社交网络数据呈现出爆炸式增长,研究多维海量社交网络数据的可视化技术具有重要意义。在分析多维和海量数据可视化技术中存在的主要问题的基础上,本文对海量社交网络数据可视化技术中的社区发现算法、力导引布局算法和多维数据可视化中的属性图聚类算法进行了重点研究。第一,针对现有社区发现算法存在社区质量不满足图可视化要求和算法效率低的问题,以及力导引布局算法存在社区结构不明显和布局效率低的问题,提出了改进的社区发现算法和社区布局算法。首先,基于Louvain算法,提出了一种面向大规模社交网络的社区发现算法。该算法结合社交网络中无尺度及小世界等特性,通过预先选取种子节点的方法,抑制了大社区的过度合并,同时也及时合并小的社区。其次,提出了一种展示大规模网络社区结构的社区布局算法。该算法通过引入社区引力促使同一社区中的节点聚拢,优化了社区引力建模,简化了算法步骤。实验结果表明,以上算法能够清晰、高效地展示海量社交网络数据。第二,针对属性图聚类算法存在社区质量不满足图可视化要求、算法效率低和具有维度灾难以及人为干预的问题,提出了改进的属性图聚类算法和基于属性映射的多维数据可视化算法。其中,改进的属性图聚类算法采用主成分分析(Principal Component Analysis,PCA)和自组织映射(Self-Organizing Map,SOM)算法分别降低节点信息维度来完成聚类。然后依据聚类结果,利用属性相似度将原有的无权网络图转换为加权网络图,之后利用本文改进的社区发现算法进行社区划分。基于属性映射的多维可视化算法以社区的角度,通过平行坐标系的方式展示数据的维度信息。以上算法具有自适应性和无监督性,能够满足大规模多维社交网络对于社区划分和可视化的要求。第三,通过集成上述改进的算法,设计了多维海量社交网络数据可视化方案,同时开发了一款可视化原型系统。
[Abstract]:With the rapid development of information technology, the social network data with multi-dimensional and magnanimous characteristics has explosive growth. It is of great significance to study the visualization technology of multidimensional mass social network data. Based on the analysis of the main problems in multidimensional and massive data visualization technology, This paper focuses on community discovery algorithm, force guidance placement algorithm and attribute map clustering algorithm in mass social network data visualization. In view of the existing community discovery algorithms, the community quality does not meet the requirements of graph visualization and the efficiency of the algorithm is low, and the community structure is not obvious and the layout efficiency is low in the force-guided placement algorithm. An improved community discovery algorithm and a community layout algorithm are proposed. Firstly, based on the Louvain algorithm, a community discovery algorithm for large-scale social networks is proposed, which combines the scale-free and small-world characteristics of social networks. By pre-selecting seed nodes, the excessive merging of large communities is restrained, and the small communities are merged in a timely manner. Secondly, In this paper, a community layout algorithm is proposed to show the community structure of a large scale network. The algorithm optimizes the community gravity modeling and simplifies the steps of the algorithm by introducing community gravity to make the nodes in the same community gather together. The experimental results show that, The above algorithms can clearly and efficiently display massive social network data. Secondly, the attribute map clustering algorithm has problems such as community quality does not meet the requirements of graph visualization, low efficiency, dimensionality disaster and human intervention. An improved attribute map clustering algorithm and a multidimensional data visualization algorithm based on attribute mapping are proposed. The improved attribute map clustering algorithm uses principal component analysis (PCA) and self-organizing map (SOM) algorithm to reduce the dimension of node information respectively. The original unauthorized network graph is transformed into a weighted network graph by using attribute similarity, and then the community is divided by using the improved community discovery algorithm in this paper. The multi-dimensional visualization algorithm based on attribute mapping is based on the community perspective. The above algorithms are self-adaptive and unsupervised, and can meet the requirements of community division and visualization in large-scale multi-dimensional social networks. By integrating the above improved algorithms, a multi-dimensional massive social network data visualization scheme is designed, and a visualization prototype system is developed.
【学位授予单位】：重庆邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP393.09;TP311.13

【参考文献】