Statistical Properties and Temporal Properties of Calling Behavior

Modern information technology makes it possible to understand, model and predict human activities by analyzing electronic traces of human behavior. In this work, we analyze more than 30 million Call Data Records (CDRs) to investigate statistical properties and temporal properties of calling behavior. We find that ranked call frequency of each number can be described by a two-stage power law function. The distribution of call duration is much different and can be described by a bimodal log-normal function. Further, we confirm that both call frequency and call duration are related with circadian cycles. During daylight, call duration is strongly correlated with call frequency. However, during evening and midnight, the pattern of call duration is much different with the pattern of call frequency. Our findings have significant impact on modeling human calling activity and optimizing telecommunication services.