Visual analytics of human mobility behavior

Abstract

Human mobility plays an important role in many domains of today’s society, such as security, logistics, transportation, urban planning, and geo-marketing. Both, government and industry thus have great interest in understanding mobility patterns and their driving social, economical, and environmental causes and effects. While stakeholders had to rely on manual traffic surveys for a long time, improvements in tracking technology made analyses based on large digital datasets possible. Recently, the omnipresence of mobile devices significantly increased the amounts of collected movement and context data. People are willing to reveal their position, but also further personal details such as visited places, observations, events, news, and sentiments in exchange for personalized services and social networking. This opens up new possibilities for many domains where a semantic mobility understanding is required but also raises major challenges. To reveal a holistic picture, heterogeneous datasets of different services with different resolution and format have to be fused and analyzed. However, social sensing data is vast, has varying scale, is unevenly distributed, and constantly updated. Especially content from social media services is often inconsistent, unreliable, and incomplete, which requires special treatment. Fully automatic mapping approaches are not trustworthy as they do not take into account these uncertainties. At the same time, manual approaches become insufficient with large amounts of data. Even when data is perfectly aligned, analysts cannot purely rely on existing techniques. Answering questions about reasons for movement requires a broader perspective that takes into account environmental and social context, the driving forces for human mobility behavior. Visual analytics is an emerging research field to tackle such challenges. It creates added value by combining the processing power and accuracy of machines with human capabilities to perceive information visually. Automatic means are used to fuse and aggregate data and to detect hidden patterns therein. Interactive visualizations allow to explore and query the data and to steer the automatic processes with domain knowledge. This increases trust in data, models, and results, which is especially important when critical decisions need to be made. The strengths of visual analytics have been shown to be particularly advantageous when problems and goals are underspecified and exploratory means are needed to discover yet unknown patterns. This thesis presents novel visual analytics approaches to derive meaning and reasons behind movement, by taking into account the aforementioned characteristics. The approaches are aligned in a holistic process model covering all steps from data retrieval, enrichment, exploration, and verification to externalization of gained knowledge for various fields of application such as electric mobility, event management, and law enforcement. It is shown how data from social media can not only be used to retrieve up-to-date movement information, but also to enrich movement trajectories from other sources with structured and unstructured information about places, events, transactions, and other observations. Through highly interactive visual interfaces analysts can bring in domain knowledge to deal with uncertainties during data fusion and to steer the subsequent semantic analysis. Exploratory and confirmatory analysis techniques are presented to create hypotheses, refine them, and find support in the data. Analysts can discover routines and abnormal behavior with assistance of automatic pattern detection methods to cope with the vast amounts of data. Spatial drill-down is supported by a set-based focus+context technique, while a more abstract visual query language allows to explicitly formulate, extract, and query for movement patterns. The approaches are applied in different scenarios and are integrated in a visual analytics system. Evaluation with experts and novice users, case studies, and comparisons to ground truth data reveal the need and effectiveness of the contributions. Overall, the thesis contributes a visual analytics process for human mobility behavior with novel semantic analysis approaches, ranging from global movements of many to local activities of a few people, for a wide range of application domains.