Semantic Enrichment of Mobile Phone Data Records Using Linked Open Data

The pervasivity of mobile phones opens an unprecedented opportunity of deepening into the human dynamics through the analysis of the data they generate. This enables a novel human-driven approach to service creation in a wide set of domains such as health-care, transportation and urban safety. The telecom operators own and manage billions of mobile network events (like the Call Detailed Records - CDR) per day: the interpretation of such a big stream of data needs a deep understanding of the context where the events have occurred. The exploitation of available background knowledge is a key element in this scenario. In this paper we introduce a novel method for the semantic interpretation of human behavior in mobility based on the merge of the mobile network data stream and the geo-referred available background knowledge. We modeled the human behavior making use of the geo and time-referenced knowledge available on the web (e.g., geo-tagged resources, info on weather forecast, social events, etc.) matching it with the mobile network coverage map. The model is intended to characterize the contexts where the mobile network events occur in order to help in interpreting the behavioral traits that generated by them. This will allow us to achieve a set of predictive tasks such as the prediction of human activities in certain contextual conditions (e.g., when an accident occurs on highway before the working time starts, etc.), or the characterization of exceptional events detected from anomalies in mobile network data. We created an ontological and stochastic high-level representation behavioral model (HRBModel) that maps the human activities to the different contexts. Given the mobile phone network and the geo-tagged resource Openstreetmap, the model is used to rank the activities associated to a particular network event (e.g. a sudden call amount peak) according to their probability. We also describe the design of an experimental evaluation and the preliminary evaluation results to measure the performance of the model and to improve the activity prediction task.