ID-CONNECT: Combining Network and Call Features to Link Different Identities of a User

The rapid growth of smart phones and recent advances in telephony have triggered an enormous increase in the number of telephony subscribers. Some people have more than one calling identity for different purposes, such as for business and personal use. Discovering multiple identities of an individual can be important for fraud detection, detection of coordinated criminals, collaborative recommendations, and analyses of complete social graph of an individual. To address the challenge of finding identities that belong to a single individual, this paper describes ID-CONNECT, an approach for linking multiple identities of an individual based on that individual's social call graph and call behavior. The approach is based on the assumption that similarity between identities can be better estimated by using both call network and call behavior of identities towards their friends and that the more similar the weighted call graph of two identities, the higher is the similarity between them. ID-CONNECT is a two-step approach: 1) it estimates similarity between identities by incorporating call rate and duration, and 2) it generates sets of candidate identities for each target identity. We experimentally evaluate our approach on different random graph models and compare its performance against other approaches. Our evaluation results show that our approach significantly reduces candidate set size (up to 70%) while achieving comparable true positive rate.