An Empirical Evaluation of Algorithms for Link Prediction

Online social networks (OSNs) analysis has been widely used in the field of information systems (IS), thus link prediction, one of the most important core techniques of OSNs analysis, plays a vital role in the development of IS. Despite the recent development of numerous link prediction approaches, there is still a lack of comprehensive studies that measure and evaluate their performance, which hinders the rational selection and full utilization of existing prediction approaches. This study proposes a novel taxonomy of link prediction approaches based on their prediction principles. Furthermore, it selects eighteen representative approaches from various categories to perform an empirical evaluation on six real-world benchmark datasets. The features of different types of predication approaches have been analyzed based evaluation test results. The research provides researchers with improved understandings on link prediction approaches and offers insightful performance related information to practitioners for developing more effective information systems.