Community detection in dynamic networks: a comprehensive and comparative review using external and internal criteria

Abstract

Context: The world is experiencing a boom of data throughout several fields such as finance, engineering, medicine, crime and security etc. More and more problems associated with these fields are starting to be modeled using networks and dynamic and/or static graphs. This has led to further exploration of these graphs, their community structures and the associated community detection methods. Objective: This paper aims at studying a set of community detection approaches and analyzing their performance in terms of both external and internal criteria, including rand index, adjusted rand index, variation of information and normalized mutual information. Method: The chosen set of algorithms is thoroughly studied. Amongst this set, each algorithm is compared with the other in the set and the result is recorded numerically. These numerical values are recorded in a tabular format for better understanding of the reader. Each table is made specific to the criterion used to compare the algorithms. Altogether, a total of 4 tables are obtained for each of the metric used for the comparative analysis. Results: On comparing the algorithms with each other, the results are tabulated in a symmetric matrix whose values along the diagonal have an associated significance, which is described in the following sections. Conclusion: This paper presents a comparative analysis of the community detection algorithms against a set of external and internal criteria. Further, it lists some challenges faced by the research community in the study of these algorithms. The last section gives a summary of the common applications of community structure and detection methods.