Data-driven Human Behavior Models: Opportunities and Challenges.

We live in a world of data, of big data, a big part of which has been generated by humans through their interactions with both the physical and digital world. A key element in the exponential growth of human behavioral data is the mobile phone. There are more mobile phones in the world as humans which has turned the mobile phone into the piece of technology with the highest levels of adoption in human history. We carry them with us all through the day (and night, in many cases), leaving digital traces of our physical interactions. Mobile phones have become sensors of human activity both in the large scale and also as the most personal devices. In my talk, I present some of the work that we are doing at Telefonica Research in the area of modeling humans from human behavioral data collected from mobile phones. The data may be data that is automatically collected by the mobile network infrastructure --in particular Call Detail Records or CDRs---or data that is collected by an experimental mobile application through a user study. All projects entail data analytics and machine learning in order to build accurate models of individual or aggregate behavior. In particular, I describe four projects: (1) a project to automatically infer personality from Call Detail Records [1]; (2) MobiScore, a project to automatically assess a credit score from Call Detail Records [2]; Borapp, a mobile app that is able to detect boredom from patterns of phone usage [3] and a project to automatically predict crime hotspots in a city from human dynamics and demographic data [4]. I conclude by highlighting opportunities and challenges associated with building data-driven models of human behavior.