@Article{cmc.2025.067783,
AUTHOR = {Hiba Sameer Saeed, Amenah Dahim Abbood},
TITLE = {A Comprehensive Review of Dynamic Community Detection: Taxonomy, Challenges, and Future Directions},
JOURNAL = {Computers, Materials \& Continua},
VOLUME = {85},
YEAR = {2025},
NUMBER = {3},
PAGES = {4375--4405},
URL = {http://www.techscience.com/cmc/v85n3/64158},
ISSN = {1546-2226},
ABSTRACT = {In recent years, the evolution of the community structure in social networks has gained significant attention. Due to the rapid and continuous evolution of real-world networks over time. This makes the process of identifying communities and tracking their topology changes challenging. To tackle these challenges, it is necessary to find efficient methodologies for analyzing the behavior patterns of dynamic communities. Several previous reviews have introduced algorithms and models for community detection. However, these methods have not been very accurate in identifying communities. Moreover, none of the reviewed papers made an apparent effort to link algorithms that can accurately detect dynamic communities. This review aims to present a taxonomy that shows several algorithms and methodologies for detecting dynamic communities. These algorithms are divided into four categories (heuristic- and modularity-based, metaheuristic, deep learning, and hybrid deep learning). It encompasses the past five years and examines the advantages and disadvantages of conventional and recent methods. Currently, many efforts are utilizing deep learning to improve dynamic networks; however, the instability of the network during the training phase affects the model’s accuracy. However, this direction remains unexplored. This study presents a review that aims to tackle this issue. We discuss a research path that explores the integration of deep learning with heuristic, metaheuristic, and hybrid metaheuristic algorithms to facilitate the identification of communities in dynamic networks. This investigation examines how this mixture surpasses the constraints of singular methodologies, resulting in enhanced detection outcomes and enabling researchers to select the most suitable algorithms for their future research.},
DOI = {10.32604/cmc.2025.067783}
}