Abstract
Online social networks, such as Facebook, have been massively growing over the past decade. Recommender algorithms are a key factor that contributes to the success of social networks. These algorithms, such as friendship recommendation algorithms, are used to suggest connections within social networks. Current friending algorithms are built to generate new friendship recommendations that are most likely to be accepted. Yet, most of them are weak connections as they do not lead to any interactions. Facebook is well known for its Friends-of-Friends approach which recommends familiar people. This approach has a higher acceptance rate but the strength of the connections, measured by interactions, is reportedly low. The accuracy of friending recommendations is, most of the time, measured by the acceptance rate. This metric, however, does not necessarily correlate with the level of interaction, i.e., how much friends do actually interact with each other. As a consequence, new metrics and friending algorithms are needed to grow the next generation of social networks in a meaningful way, i.e., in a way that actually leads to higher levels of social interactions instead of merely growing the number of edges. In this paper, we develop a novel approach to build friendship recommender algorithms for the next-generation social networks. We first investigate existing recommender systems and their limitations. We also highlight the side effects of generating easily accepted but weak connections between people. To overcome the limitations of current friending algorithms, we develop a clustering-based interaction-driven friendship recommender algorithm and show through extensive experiments that it does generate friendship recommendations that have a higher probability of leading to interactions between users than existing friending algorithms.