微博用户模型复杂网络中多维有向社区发现

MULTI-DIMENSIONAL DIRECTED COMMUNITY DETECTION IN COMPLEX NETWORK OF MICROBLOGGING USER MODEL

大多数社区发现是基于一种信息的,即从一个维度来划分社区。但在现实场景中,用户之间社区构成是受兴趣、社交关系、地域、教育背景等诸多因素共同影响形成的。这些多维信息有些是无向的,如兴趣相似度等;有些是有向的,如关注关系等。根据有向社区发现的原理,将多个维度的信息融合,提出一种面向多维复杂网络的有向社区发现(MDCD)算法。通过实验证明,MDCD算法相对于传统的多维社区发现方法 AMM算法,社区发现结果准确率提高了17.7%、F-measure值提高了0.068;与一维的兴趣相似度网络进行对比,MDCD算法的三维复杂网络社区发现结果的准确率提高了36.1%、召回率提高了25.3%。由于多维有向社区发现综合考虑了多维的信息,得到的社区结构具有更重要的社会意义。

Most of community detections are based on one kind of information,i. e. to partition the community using one dimension. However in reality scene,the composition of communities between users is formed by the combined effect of many factors,such as interests,social relationships,geography,educational background,etc. Moreover,some of these multi-dimensional information are undirected,for ex.,the similarity of interests,but some others,like the relationship of follow,are directed. Based on the principle of directed community detection,in this paper we fuse the multi-dimensional information and propose a multi-dimensional complex network-oriented directed community detection algorithm( MDCD). It is proved through experiment that the MDCD algorithm,relative to conventional multi-dimensional community discovery algorithm AMM,improves the accuracy of community detection result by 17. 7% and the F-measure value by 0. 068; Furthermore,by comparing the MDCD algorithm with the one-dimensional interests similarity network,it improves the precision rate of three-dimensional complex network detection result by 36. 1% and the recall rate by 25. 3%. Since the multi-dimensional directed community detection considers the multi-dimensional information comprehensively,the community structure obtained has more important social significance.