WCSE 2016
ISBN: 978-981-11-0008-6 DOI: 10.18178/wcse.2016.06.044

Link Prediction’s Best Value Interval

Jia Jun, Hu Xiaofeng

Abstract— We introduced three new accuracy metrics which are Dynamic Precision, Prediction Rate and Prediction Efficiency at first and then we did the link prediction experiments with four real networks which are USAir, NetScience, Jazz and metabolic. After analysis the experiments’ results, we found that most missing edges of different networks in different prediction results gathered in the similar interval which is the top one fifth of the total number of the original network’s edges. We also found that the accuracy of link prediction could hardly get a high value when the missing edges’ ratio is too few and it will drop sharply when the ratio is too large. These conclusions could provide basically reference for link prediction’s practical application in future.

Index Terms— link prediction, accuracy metrics, dynamic precision, best value interval.

Jia Jun, Hu Xiaofeng
National Defense University, CHINA