The large-scale understanding of personal and social behavior from mobile sensor data has emerged as a frontier domain in computing. Mobile phones, constantly sensing people’s location, motion, communication, and proximity to others, represent the most accurate and non-intrusive current means of tracing human activities. All this information, as never before, is being generated at massive scales. With appropriate privacy-protection mechanisms in place, the use of mobile phones as input to automatic learning of people’s routines and relations opens the possibility to a multitude of social computing applications and services, including personalized
information access, sensible healthcare, personal self-assessment, and creative expression.
We propose to develop new algorithms that, robustly integrating mobile sensor data captured from people’s daily life, are able to automatically discover personal and social context, including personal routines, group relationships, and group routines from large-scale observations, both longitudinally and population-wise. Our methods aims at responding to queries like the following: What are the most common daily or weekly routines for a given phone user? What are the most likely places for a user on a given day? Who are the people a user will most likely meet tomorrow morning? What are the existing communities in a given population, and how are they related? Which are the likely places and times in which a user meets her communities? Was today an unusual day for a user, in terms of his personal typical patterns?
We propose to reach these goals through three fundamental research objectives:
1. Development of novel algorithms for representation of human context at the personal and group level from raw sensor data, including location, motion, proximity, and communication, which are capable of integrating heterogeneous sources of sensor data, are robust enough to handle the challenges associated with noise levels, sparseness, and discrimination power of mobile sensor data, and are accurate enough to provide measurements of the physical and social pace of people’s lives.
2. Development of new methods for discovery of personal and social context from instantaneous and aggregated data representations, based on the design of advanced, principled probabilistic models for discovery of personal routines, i.e. regularities in people’s lives, possibly over different time scales, and discovery and characterization of groups from direct communication patterns, mutual proximity, and similar personal routines, that can be used as starting points for future context-based mobile services.
3. Creation of research resources for large-scale human context modeling from cell phone data, including a carefully designed, privacy-sensitive data collection campaign using state-of-the art Nokia phones, and intended to be open to the research community, and the development of the infrastructure needed to record, manage, and demonstrate both the collected sensor data and the results of the performed research.
Overall, we believe that this project represents a unique and timely opportunity to build on the scientific and technical expertise on perceptual and social computing available at Idiap, and the strategic objectives and expertise at PeC Lab and Nokia at large.