Abstract
Most of the existing algorithms proposed for communities determination are based on the topological features of social networks. They do not take into account information shared between users. In this paper, We introduce a graph mining algorithm to detect social network communities a MeanCD (meaningful community detection). The proposed algorithm encloses a function measuring the strength of links describing the information (topics) shared and friends and more generally between users. This information is called Link-importance. It is based on two main steps. The first one consists in defining the community after cleaning the networks, while the second step is to determine the good partitioning which maximizes the new concept of the quality function called Modularity.