Community detection in multi-relational directional networks

Abstract

Community detection algorithms are essential tools that allow us to discover organizational principles in networks. Today, in most cases, social networks are modeled as a multi-relational network. So far, despite the introduction of community detection algorithms specific to multi-relational networks, very few of these algorithms have been able to find overlapping communities in a multi-relational directional network. In this paper, we present OCMRN (Overlapping Communities in Multi-Relational Networks), an overlapping community detection method in multi-relational directional networks. A semi-supervised method is used in the training phase of this algorithm which allows determining the importance of each layer to shape communities. The proposed method is evaluated on nine real and eight synthetic datasets and is compared with different algorithms. Well-known evaluation criteria are used to compare overlapping and non-overlapping communities resulting from various algorithms. The high accuracy of the results obtained from the proposed model suggests this algorithm can be used with high confidence for community detection of multi-relational directional networks to find overlapping communities.