Common Features against Similarity for Discovering Social Circles in Networks

This paper presents a new method for automatically identifying circles of friends in ego networks. To detect these circles the users' network is investigated in two different ways: (1) based on features that the users own, and (2) based on similarity between the central user and his friends. In both approaches, the user profile information is taken into account. The overlapping and nested circles are also detected. The algorithm described here successfully achieved a high matching rate between generated and hand-labeled real-world circles, albeit the best results were achieved only with a large number of generated circles. The experiments showed that both developed algorithms produced distinctly better results than the already known methods. Further, the former based on common features gives better results than the latter based on similarity.