f-Divergence Measures for Evaluation in Community Detection

Mariam Haroutunian (1) , Karen Mkhitaryan (1) , Josiane Mothe (2, 3)

1 Institute for Informatics and Automation Problems
2 IRIT-SIG - Systèmes d’Informations Généralisées
3 UT2J - Université Toulouse - Jean Jaurès

Community detection is a research area from network science dealing with the investigation of complex networks such as biological, social and computer networks, aiming to identify subgroups (communities) of entities (nodes) that are more closely related to each other than with remaining entities in the network [1]. Various community detection algorithms are used in the literature however the evaluation of their derived community structure is a challenging task due to varying results on different networks. In searching good community detection algorithms diverse comparison measures are used actively [2]. Information theoretic measures form a fundamental class in this discipline and have recently received increasing interest [3]. In this paper we first mention the usual evaluation measures used for community detection evaluation We then review the properties of f -divergence measures and propose the ones that can serve community detection evaluation. Preliminary experiments show the advantage of these measures in the case of large number of communities.

Mots clés: Community detection; Information theory; f-divergence distance measures

Domaines: Théorie de l'information [cs.IT]

Mariam Haroutunian, Karen Mkhitaryan, Josiane Mothe. f-Divergence Measures for Evaluation in Community Detection. Workshop on Collaborative Technologies and Data Science in Smart City Applications (CODASSCA 2018), Sep 2018, Yerevan, Armenia. pp.137-145. ⟨hal-02280352⟩