During the last 15 years there has been an explosion in human behavioral data caused by the
emergence of cheap electronics and online platforms. This has spawned a whole new research
field called computational social science, which has a quantitative approach to the study of
human behavior. Most studies have considered data sets with just one behavioral variable such
as email communication. The Social Fabric interdisciplinary research project is an attempt
to collect a more complete data set on human behavior by providing 1000 smartphones with
pre-installed data collection software to students at the Technical University of Denmark.
The data set includes face-to-face interaction (Bluetooth), communication (calls and texts),
mobility (GPS), social network (Facebook), and general background information including a
psychological profile (questionnaire).
This thesis presents my work on the Social Fabric data set, along with work on other
behavioral data. The overall goal is to contribute to a quantitative understanding of human
behavior using big data and mathematical models. Central to the thesis is the determination of
the predictability of different human activities. Upper limits are derived for communication,
movement, and social proximity, along with upper limits on the predictability of the spatial
dynamics. Furthermore, it is shown that different human activities have a strong impact on
each other and that human activity patterns follow a general linear principle throughout the
population.
It is found that the type and strength of an interaction impacts the tendency for the
interaction to take place in a relation formed by two similar individuals. It is furthermore
shown that most relations are asymmetric in the sense that one part initiates communication
much more often than the other. Evidence is provided, which implies that the asymmetry is
caused by a self-enhancement in the initiation dynamics. These results have implications for
the formation of social networks and the dynamics of the links.
It is shown that the Big Five Inventory (BFI) representing a psychological profile only
correlates weakly with the behavioral variables collected by the smartphone. In particular,
the BFI almost does not impact our choice of friends, our interaction patterns, and our mobility
patterns. The extraversion variable provides the only exception to this rule.
The dynamics of online attention on Twitter is studied and the underlying stochastic
dynamics is derived under a Markovian assumption. Finally, it is studied how the occurrence
of new aircraft incidents influence the memory of old aircraft incidents on Wikipedia. New
incidents are found to trigger a large flow of attention to old incidents, thereby implying that
interactions between spreading processes are driving forces of attention dynamics.
Overall, the thesis contributes to a quantitative understanding of a wide range of different
human behaviors by applying mathematical modeling to behavioral data. There can be no
doubt that such mathematical models can have great predictive power, but obviously there
are also limitations.