Relational similarity model for suggesting friends in online social networks

Abstract

Suggesting friends is a very important aspect in any online social network. In this paper, we present a relational similarity model for suggesting friends in online social networks, which uses relational features as opposed to the non-relational features that are used in current friend suggestion applications. We take a supervised learning approach and build a model that uses information of not only the two central users but also of their current neighborhoods. We use a dataset from Facebook to evaluate the accuracy of our model by comparing the performance of feature sets belonging to relational/non-relational categories and boolean and numerical sub categories. We show experimentally that the relational information improves the accuracy of boolean features but does not affect the performance of numerical features. Moreover, we show that our overall model is highly accurate in recommending people in online social networks.