Social Circle-Based Algorithm for Friend Recommendation in Online Social Networks

Recommending friends to registered users is a crucial personal service of Online Social Networks(OSN).OSN will recommend a friend to a user if they share some common attributes or neighbors.But the recommendation accuracy is usually not so good since users' profile information may be incomplete and the relationships between neighbors are ignored.In fact,users can group their friends into several social circles and two users are more likely to become friends if they share similar social circles.Therefore,a social circle detection algorithm is suggested at first,and then the social circle similarity is defined.Based on this similarity,we can recommend friends to a user.Our hypothesis is verified by statistically analyzing the YouTube dataset.To verify the efficiency of the social circle detection algorithm,the ego networks of Facebook are used.The experimental results show that compared with three typical detection methods,our approach can identify social circles efficiently and accurately.We utilize social circle similarity, common neighbor similarity and Jaccard similarity to predict friend relationships in Facebook New Orleans network.The experimental results provide strong evidence that our algorithm is more precise in friend recommendation.