Data Mining and Visualization of Large Human Behavior Data Sets

Andrea Cuttone

Research output: Book/ReportPh.D. thesisResearch

3299 Downloads (Pure)

Abstract

Traditional methods for studying human behavior such as surveys and manual collection are expensive, time-consuming and therefore cannot be easily applied at large scale. In recent years an explosive amount of digital traces of human activity – for example social network interactions, emails and credit card transactions – have provided us new sources for studying our behavior. In particular smartphones have emerged as new tools for collecting data about human activity, thanks to their sensing capabilities and their ubiquity. This thesis investigates the question of what we can learn about human behavior from this rich and pervasive mobile sensing data. In the first part, we describe a large-scale data collection deployment collecting high-resolution data for over 800 students at the Technical University of Denmark using smartphones, including location, social proximity, calls and SMS. We provide an overview of the technical infrastructure, the experimental design, and the privacy measures. The second part investigates the usage of this mobile sensing data for understanding personal behavior. We describe two large-scale user studies on the deployment of self-tracking apps, in order to understand the patterns of usage and non-usage. Moreover we provide some design guidelines for facilitating reflection in self-tracking systems. Finally we propose a model for inferring sleep patterns from smartphone interactions.

In the third part, we focus on a specific aspect of collective behavior: human mobility. We perform an experiment to verify the feasibility of inferring places from location traces using mobile sensing data. We develop a hierarchical model for human mobility, which is able to measure mobility properties at multiple scales. We perform a study on the factors influencing the accuracy of nextplace prediction models. Finally we present an open-source tool for creating geographical visualizations.
Original languageEnglish
Place of PublicationKgs. Lyngby
PublisherTechnical University of Denmark
Number of pages258
Publication statusPublished - 2017
SeriesDTU Compute PHD-2016
Number422
ISSN0909-3192

Fingerprint

Dive into the research topics of 'Data Mining and Visualization of Large Human Behavior Data Sets'. Together they form a unique fingerprint.

Cite this