A recommender system based on local topology features and message rate

Social networks have become an indispensable part of every day's life. A vast majority of people use at least one social network to communicate with friends or business partners. Different applications of social networks are used to meet diverse needs of users. Social networks use recommender systems to provide wider experience for their users. Friend recommendation has been the most popular and important application used to expand the circle of friends and social communication. Social networks are very dynamic. Every moment new people are added and new forms of relationship are formed. Therefore, evaluating social networks is a complicated task. In this paper, the focus is both on the characteristics of social networks as well as the messages sent between and among the nodes. A mixed method will be suggested which will initially use a community detection algorithm based on message rate as a pruning algorithm. It will then divide the network into several communities. Afterwards, the FriendLink algorithm will be employed in order to estimate the similarity score between all users based on the properties of the network structure and the rate of the messages sent. Finally, users with maximal similarity will be suggested as friends to the target user. The results of testing a Facebook-like dataset revealed that the suggested method is of a great precision and accuracy as compared to FriendLink and CSM algorithms.