inis 21(25): e4

Research Article

Structural Importance-based Link Prediction Techniques in Social Network

Download51 downloads
  • @ARTICLE{10.4108/eai.7-1-2021.167840,
        author={Abdul Samad and Muhammad Azam and Mamoona Qadir},
        title={Structural Importance-based Link Prediction Techniques in Social Network},
        journal={EAI Endorsed Transactions on Industrial Networks and Intelligent Systems},
        volume={7},
        number={25},
        publisher={EAI},
        journal_a={INIS},
        year={2021},
        month={1},
        keywords={Link Prediction, Social Network Analysis, Similarity Measure, Structural Importance, Centralization},
        doi={10.4108/eai.7-1-2021.167840}
    }
    
  • Abdul Samad
    Muhammad Azam
    Mamoona Qadir
    Year: 2021
    Structural Importance-based Link Prediction Techniques in Social Network
    INIS
    EAI
    DOI: 10.4108/eai.7-1-2021.167840
Abdul Samad1,*, Muhammad Azam2, Mamoona Qadir3
  • 1: Capital University of Science and Technology, Islamabad Pakistan
  • 2: The University of Agriculture Faisalabad
  • 3: Khwaja Fareed University of Engineering and Information Technology, Rahim Yar Khan Pakistan
*Contact email: writetosamadalvi@gmail.com

Abstract

Link prediction in social network gaining high attention of researchers nowadays due to the rush of users towards social network. Link prediction is known as the prediction of missing or unobserved link, i.e., new interaction is going to be occurring in a near future. State-of-the-art link prediction techniques (e.g., Jaccard Index, Resource Allocation, SAM Similarity, Sorensen Index, Salton Cosine, Hub Depressed Index and Parameter-Dependent) considers only similarity of the pair of node in order to find the link. However, we argued that nodes having same status of centralization along with high similarity can connect to each other in a future. In this paper, we have proposed structural importance-based state-of-the-art link prediction techniques and compared. We have compared structural importance-based link prediction techniques with state-of-the-art techniques. The experiments are performed on four different datasets (i.e., Astro, CondMat, HepPh and HepTh). Our results show that structural importance-based link prediction techniques outperformed than state-of-the-art link prediction techniques by getting 95% at threshold 0.1 and 68% at threshold 0.7.