Performance Analysis of a Multicore Approach Proposed for Efficient Community Detection and Recommendation System

Abstract

Community detection is a well-known area of research that yields essential results in various fields such as social media, biological networks, pandemic spread, recommendation systems, etc. There are two important areas for improvement in existing community detection algorithms: quality of community formed should be improved and a parallel approach to community detection is needed to handle massive data. In this paper, we have proposed a three-step parallel algorithm, Par-Com, using the concept of k Clique and modularity optimization to address both of the above issues. The proposed algorithm increases the execution speed and also improves the quality of the community formed by optimizing modularity. Par-Com uses dynamic load balancing on the multicore architecture of Supercomputer ParamShivay. We have also evaluated Par-Com’s performance against nine sequential and three parallel community detection algorithms on varying size datasets, i.e. karate, macaque, email, immuno, soc-epinions, facebook, and com-Friendster. The experiment result shows that Par-Com outperforms other algorithms under consideration with up to 45% increase in modularity and up to 84% increase in execution speed. Par-Com is also capable of detecting overlapping communities, fuzzy membership of each node, most influential node in each community formed, and outlier nodes. Nodes within a community that have the highest influence are deemed as experts. The choices made by an expert in a particular community are served as recommendations to other users within that community.

Keywords:

• Community detection
• Modularity optimization
• Overlapping community
• Parallel computing

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by University Grants Commission.

Notes on contributors

Dipika Singh

Dipika Singh received MCA degree from RGPV University, Bhopal in 2011; qualified GATE in computer science and information technology (2012) with 2135 rank; qualified UGC-National Eligibility Test (NET) in computer science and applications (March 2013); qualified GATE in computer science and information technology (2013); awarded junior research fellowship (July 2018) and subsequently senior research fellowship by the University Grants Commission (UGC), New Delhi, India (2017–2022). Selected as lecturer, computer science in UPPSC Polytechnic, 2021 (Gazetted).

Rakhi Garg

Rakhi Garg is associate professor in Department of Computer Science, MMV, BHU; qualified UGC-NET for eligibility for lectureship in computer science & applications in universities/institutions throughout the country. Done PhD in association rule mining algorithms at Department of Computer Science, Institute of Science, Banaras Hindu University, Varanasi (India). Earlier, She completed master's in computer science from BHU, Varanasi, India in 1997 and bachelor's degree from the same institute in 1995.Email: [email protected]