Multiobjective evolutionary algorithm with graphlet measure for detection of complex topological and biological community structure

• Main Author: Abduljabbar, Dhuha Abdulhadi
• Format: Thesis
• Language: English
• Published: 2021
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://eprints.utm.my/id/eprint/101493/1/DhuhaAbdulhadiAbduljabbarPSC2021.pdf.pdf; http://eprints.utm.my/id/eprint/101493/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:150558
• Institution: Universiti Teknologi Malaysia

In the past few years, community structure detection has garnered much attention due to its significant role in analyzing complex network structures and functions. Detecting natural divisions in complex networks is proved to be an extremely Non-deterministic Polynomial-time hard (NP-hard) problem, which has been solved using evolutionary computation methods. Despite many efforts to design an effective community structure formula, the definition is still general, depending solely on the nodes' intra- and inter-connections. It lacks complete reflection of inherent topological properties, such as graphlet measure in terms of graphlet degree signatures and signature similarities, that can accurately detect complex communities' structure such as topological and biological community. The research proposes a new method termed MOEA_CGN (MultiObjective Evolutionary Algorithm based on Cooperation between Graphlet-based measure and Neighborhood relations) to improve the detection quality of complex topological and biological community structure in terms of accuracy and velocity. Thus, the contribution of this study is summarized in threefold. First, a new multiobjective optimization function is proposed to tackle the issue of a community structure definition. Second, a heuristic mutation operator is designed to enhance MOEA_CGN performance to accurately detect complex topological community structure by tackling the resolution limit problem. Third, the heuristic mutation operator is improved to make the MOEA_CGN method identify and detect complex biological community structure accurately by tackling the heterogeneity issue of real-world networks. Systematic experiments on different realworld networks from various domains and synthetic networks of different complexities have demonstrated the proposed method's effectiveness and robustness to define the community detection problem. Specifically, the proposed method has achieved detection reliability with an average improvement of 6.7% in detecting complex topological communities and 9.17% in detecting complex biological communities compared with the state-of-the-art benchmark studies. Moreover, the proposed MOEA_CGN method has demonstrated its ability to detect the optimal community structures in most complex networks faster than the competent multiobjective models based on community detection.