Trajectory Mapping in Physical and Topical Spaces

In this dissertation, two different types of trajectories are investigated. In the engfirst part of this work, we investigate methods for the analysis of physical trajectories. We focus on scenarios, in which signal strengths ofWiFi access points are recorded. In detail, they are recorded through smartphone devices by people moving through a building. By means of theWiFi signal strengths, (physical) trajectories of humans are reconstructed, which reflect the locations visited over time. Accurate localization based onWiFi signal strenghts is difficult due to different disturbances and differences in sensor hardware. Hence, our aim is not to create an accurate floor plan, but locations and trajectories in this work are reflected by an automatically constructed topological map, which represents locations and some relation between them (e.g., being visited one after another). In the second part of this thesis, we investigate scientific trajectories in topical spaces. Similar to any person leaving trajectories in physical spaces, scientists leave trajectories in topical spaces by means of the publications they write. The research of scientific authors is often focussed on one or several specific (research) topics. Over the years, these focussed topics may change, resulting in what we - in this work call - topic space trajectories. Next to authors, trajectories in topic space can also be created by scientific venues, i.e., conferences or journals. This may reflect characteristics of the venues. On a larger scale, trajectories may also give an overview over the overall changing interest in scientific topics. We investigate, in this work, methods for mapping trajectories of both, scientific authors and venues, reaching for methods such as dimension reduction and topic models. Based on these approaches, we also investigate how topical expertise is passed between scientists as topic flows through co-author networks. Finally, we investigate a practical scenario in which the topics of scientific venues are used to create an explainable, scientific venue recommendation system. The different aspects of our work are connected through a case study on the different approaches. In the physical trajectory setting, our case study is comprised of, first, an office scenario in which people follow their normal day behavior and, second, an exhibition scenario in which different booths distributed within a multifloor building are visited during a congress. For topical trajectories, we conduct the case study on top machine learning conferences and journals as well as important authors from this field. Finally, for topic flows, the case study is continued on large publication data sets from mathematics and computer science, comprised of about 20 Mio. publications spanning over 60 years of research. The case study demonstrates the benefits of our approaches.