A Label Propagation-Based Algorithm for Community Discovery in Online Social Networks.

With the rapid development of Internet and Web 2.0 applications, many different patterns of online social networks become fashionable all over the world. These sites help people share and exchange information, as well as maintain their social relations on the Internet. Therefore, it is very important to study the structure of communities in online social network. Most of existed community discovery algorithms are very costly. Moreover, the behavior of users in online social networks is rather dynamic. We first investigate Label Propagation Algorithm (LPA), which has near linear time complexity and discuss some limitations of LPA. Then, we propose a new algorithm for community discovery based on label influence vector (LIVB), an improved variation of LPA. In this algorithm, we abstract several types of nodes corresponding to different kinds of entities such as users, posts, videos as well as comments. Different types of relations between nodes are also taken into account. A node will update its label by calculating its label influence vector.We conduct experiments on crawled real data and the experimental results show that communities discovered by LIVB algorithm have more concentrative topics. The quality of the communities is improved and LIVB algorithm remains a near linear time complexity. © Springer-Verlag 2013.