Mining Friendship from Cell-Phone Switch Data

Identification of social relationships in social networks, especially cell phone networks, has attracted a lot of attention recently. Mobile phones continue to be one of the main methods of communication. Their voice and SMS logs are accessible in network switches and/or telecommunication centers. However, relationships like friendship have been mined from several features that are hard or impossible to obtain like proximity. With this in mind, in the following study, we are analyzing the voice and SMS logs of ´Reality mining´ dataset without any hard to obtain attribute to predict different relationships among the study participants with just one factor. The dataset, which encloses communication logs in addition to some other information about subjects, is focused on 90 people at MIT Media Lab from 2004 to 2005. We present an algorithm based on just one factor to infer type of friendship among study participants. The authenticity of our analysis is proven by comparing the results of our analysis with self-reported surveys subjects provided regarding their relationships. Although we performed our analysis without any prior knowledge or hard to obtain feature on ´Reality Mining´ participants, we have achieved almost the same accuracy and detection rate compared to previous studies using all the features.