Social relationship classification based on interaction data from smartphones

Wireless Communications and Mobile Computing have fundamentally changed the way people interact and communicate with each other. As the command-center of the user's communications with the outside world, smartphones hold the key to understand the user's social relationship with other people of interest. In this paper, we propose to use the unique multi-model interaction data from smartphone to classify social relationships. We firstly carry out a social interaction data collection campaign with a group of smartphone users to obtain real-life multi-modal communication data and model the data as a social interaction matrix. Then we perform a statistical analysis on the social interaction matrix to identify the interesting interaction patterns in the data. After applying different classification algorithms on social interaction matrix, we find that SVM outperforms KNN and decision tree algorithms, with a classification accuracy of 82.4% (the accuracies of KNN and decision tree are 79.9% and 77.6% respectively). Additionally, with dimensionality reduction algorithms, we embed the social interaction data containing 65 features into a 9-dimensional space while preserving the high classification accuracy. We also demonstrate the viability of applying CUR decomposition to identify important features so as to conserve energy during interaction data collection. In particular, based on the 13 out of 65 features selected by the CUR approach, we can still achieve classification accuracy of 77.7% while substantially cut down the amount of raw interaction data to be collected, stored and processed.