link age: a factor in link prediction in a social network

Open Access
Akcay, Samet
Graduate Program:
Electrical Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
May 11, 2015
Committee Members:
  • David Jonathan Miller, Thesis Advisor
  • Social Network
  • Link Prediction
  • Prediction Power
  • Network Theory
  • Mixture Model
  • Logistic Regression
This work extends a previous one that investigated link age and its effect on network evolu- tion. Whether aging adversely influences prediction power of links in network evolution is the fundamental question partially answered in the previous work. Additionally, this study argues whether reliable old connections in a network have a great impact on future link predictions. One of our hypotheses is that aging of a link is a crucial factor in link prediction. The other one is that prediction power of a link usually lessens over time. Using logistic regression and mixture extension, younger links are observed to dominate the link prediction process in most cases. However, this is not always the case. We cannot ignore the links that are old but still powerful. In addition to prediction power of the links, using a mixture model improves the overall link prediction accuracy. The findings of this research support the implications of the previous work that some old and unstable links might be removed from the network.