Mining Social Signals in Cyber-Human Systems: Collective Behavior, Personal Health, and Modeling Methods

With increasingly advanced pervasive technology many applications centered on assessing human behavior and health stand to benefit from new data and analytics. Among the new data captured by smart technology, social signals, the stimuli exchanged via modes of online and offline social interactions, are promising yet under-exploited source of information to help understand and infer human outcomes. This dissertation is focused on the data mining methodologies that transform social signals data available from smart devices in daily use into human serving insights. Specifically, I focus on three major components: (1) using Twitter data and protest participation theory to forecast daily civil unrest activities during the Arab Spring, through which I demonstrate the value of theoretical underpinnings in mining online social signals for macro-level, collective behavior prediction; (2) using smartphone-based physical proximity data to improve cognitive stress recognition through two novel feature engineering methods that are applicable to generic social signals for micro-level, personal outcome inference, and; (3) the theoretical connections between inverse reinforcement learning and relational event model in discovering group social interaction dynamics, through which I broaden the scope of modeling methods for characterizing sequence of social signals in cyber-human systems. I then propose future research directions incorporating and integrating multiple sources of human-centered sensing data to contribute to aspects of personal well-being and collective good.