TY  - JOUR
AU  - Patidar, Vinod 
AU  - Patil, Jayashree A. 
AU  - Thapak, Sonika 
AU  - Singh, Manmohan 
AU  - Prasad, Vikas 
AU  - Ayyub, Shaheen 
AU  - Sharma, Dharmendra 
PY  - 2024
TI  - Personalized Recommendation System: Web of Things Using Modular Density-Based Community Discovery
JF  - Journal of Computer Science
VL  - 20
IS  - 9
DO  - 10.3844/jcssp.2024.918.930
UR  - https://thescipub.com/abstract/jcssp.2024.918.930
AB  - Community discovery is the cornerstone and core of study in customized recommendation, assembling group features and social network analysis on the web of things. Conventional community discovery methods, however, struggle with difficulties including low accuracy, delayed convergence, modularity resolution limits and more when dealing with more complicated social networks. Because of this, differential evolution and module density are included in community discovery and a better differential evolution and module density community discovery approach is offered. The method first alters the mutation strategy and differential evolution parameters and then uses the module density as a fitness function to get beyond the restriction of modularity resolution improve population quality overall and hasten the process of global convergence. Experiments with various commonly used community discovery techniques using computer-generated network datasets and sample real-world network datasets. When Collective Co-Evolutionary Differential Evolution-based Community Detection (CCDECD) and Classification-based Differential Evolution algorithm for Modularity Optimization (CDEMO) are used simultaneously, the difference is optimal and the Q value increases by 3.3% compared to the Overlapping Community Detection algorithm based on Density Peaks (OCDDP) and 4.6% compared to GN. However, the mutual information value NMI is not optimal since the GN method is better suited for small-scale networks. The division result of Improved Differential Evolution and modularity density Community Detection (IMDECD) on the network is the closest to the actual network since the NMI is ideal, the Q value is low and the standard deviation is minimal. The suggested method has improved accuracy and better convergence performance, according to the experimental findings.