Evaluations of Similarity Measures on VK for Link Prediction

Abstract

Recommender system is one of the most important components for many companies and social networks such as Facebook and YouTube. A recommendation system consists of algorithms which allow to predict and recommend friends or products. This paper studies to facilitate finding like-minded people with same interests in social networks. In our research, we used real data from the most popular social network in Russia, VK (Vkontakte). The study is motivated on the assumption that similarity breeds connection. We evaluate well-known similarity measures in the field on our collected VK datasets and find limited performance results. The result shows that majority of users in VK tend not to add possible users with whom they have common acquaintances. We also propose a topology-based similarity measure to predict future friends. Then, we compare our results with the results of other well-known methods and discuss differences.