The visualization and analysis of large graphs plays an essential role in various application fields. Since the size of graphs grew exponentially in the past few years, it became a challenge to reduce the visual clutter of dense and occluded graphs. By abstracting the structure of a node-link diagram, containing thousands of nodes and edges, visual clutter is reduced drastically, supporting the analysis of underlying patterns in an interactive approach. Additional visual techniques are used to overcome the challenge of representing the evolution of structural diagram changes and relationships between entities in dynamic graph visualization. The recent publications of large static and dynamic graph visualization techniques are using rich clients based on fast processing GPU algorithms, as well as distributed approaches for cluster-computing frameworks. Even though these techniques are capable of processing large-scale graphs interactively, they are also restricted by the user’s hardware or are more complex and expensive than simple client-server solutions. This thesis aims to provide an alternative approach, at providing a distributed, cross-platform, server-client application, able to visualize large node-link graphs, consisting of thousands of elements, interactively in a standard web-browser. We describe an aggregation strategy based on meta-elements, that provides an adjustable level of detail interface and visualizes the hierarchy of cumulative elements throughout multiple abstraction layers. By highlighting structural changes over time in dynamic graphs in combination with tools, such as panning and zooming and overview and detail, our system allows for dynamic graph exploration. We will demonstrate the usability of our technique by providing a complete prototype and present benchmarks on different graphs. Furthermore, we evaluate technical aspects of our approach as well as its applicability to large real-world graphs.
Visual analytics
The digitization of our world provides us with a vast amount of data. This data allows us to construct accurate models of real world situations which are explored and analyzed to get a deeper understanding and eventually draw conclusions for our further actions. Multivariate networks are a particularly complex construct which are ubiquitous in many different subject areas, like social media, telecommunication, transport, finance, and demographics. These networks often have a spatial context attached to them and usually evolve over time. This fact makes it even harder to efficiently visualize the many aspects of such a network. This thesis aims to define and build a visualization of a multivariate network which changes over time and space. The underlying data network is composed of real-world movement data of citizens of Vienna from 2007 to 2018, provided by the city of Vienna, MA23. This data represents the change of residencies of people moving to, from, or within Vienna. To tackle the complexity of the many dimensions of this data such as time, space, and other attributes, like the country of birth of the moving people, we follow a user-centered design approach proposed by Miksch et al. The implemented prototype of the visualization focuses on two different user groups, which are people from the department for urban development on the one hand and the public on the other hand. Both groups may take interest in the relations between the districts and in understanding the migration flow over the years. In the design process, we focus on strengths and weaknesses of different visualization techniques to amplify the visual expressiveness of the key aspects of the data. Spatial information is encoded in a geographic map on which flows depict movements between areas. The design choices of these flows are essential to sustain readability. The temporal aspects are depicted with different time-series visualizations. Each of them focuses on the data from a different angle. Interactivity and interoperability between these visualizations ensure determined navigation through the various aspects of the migration data. We evaluated the visualization prototype with five experts in the field of Visual Analytics and one non-expert. The evaluation showed that the right combination of different visualization and interaction techniques results in an effective and appropriate visualization from which users can draw the desired insight.
The research area "Knowledge-Assisted Visual Analytics" (KAVA) deals with the integration of domain knowledge into Visual Analytics approaches which offers many advantages for research as well as for the analysis of data since analysts do not need to rely on their domain knowledge and can concentrate more on the analysis task itself. Especially in the health care sector, KAVA has great potential which is currently not fully exploited since there are only a few approaches that deal with KAVA in combination with health care data. To fill this gap, we propose a new KAVA approach dealing with a dataset that resulted from a clinical trial of a medication for treating the eye disease Uveitis to provide the possibility of exploring and analyzing the dataset. For designing and developing the approach a user-centered design process, involving a domain expert, in combination with problem-driven visualization research is applied. The final approach is validated using a qualitative task-oriented user study with five visualization experts. The results suggest that the approach is able to support the analysis as well as exploration of the dataset.
With the increase of remote work due to COVID-19 and the overall movement towards open source projects, distributed version control system, like Git gained popularity overthe last years. The publicly available data on platforms (e.g., GitHub) therefore becomes richer and attracts sociologists and software analysts for further analysis.This master thesis aims to visualize GitHub trends using Visual Analytics. The data used originates from the GitHub API as well as GitHub Archive, is multivariate and contains different types of information containing repositories, users and events. This data will be extended by the temporal dimension to identify potential trends. For the problem definition and further methodology, the design triangle as described by Mikschet. al is being used.The outcome of the thesis is a prototype, that not only enables domain experts to fulfill common tasks related to identifying GitHub anomalies and trends but also allows foruser interaction to focus on more granular analysis. While many trends can potentially be visualized, this thesis will focus on a small subset of trends to introduce a generic approach and evaluate it on given scenarios and tasks. The general group of potential user groups is broad, but there is a strong emphasis on analysts in technology industries.The prototype was evaluated with domain experts in different fields of expertise that were asked to perform given tasks that can be fulfilled using the developed prototype. The results of the evaluation showed, that there is a strong interest in the analysis of GitHub data and that the right encodings and visualization methods can help find patterns and trends significantly.
The global expansion of wind energy requires robust and meaningful geographic information about its locations. Studies have shown how enriched global data on wind infrastructure can be generated using OpenStreetMap but have neglected to represent and make it analyzable using visual tools. For an accurate visual investigation, knowing which parameters can be used to characterize wind farms and which visual encodings are suitable for global and local analysis are essential. With this aim in mind, we conducted a design study that produced a dataset called the Enriched Data of Wind Farms (EDWin) and a prototype for its interactive visualization. Through a user study, we evaluated the tool's appropriateness for exploring unproven claims about wind farms from the literature and identifying specific wind farms characteristics through simplified visual encoding. The prototype enabled users to complete the tasks, but many needed help from the interviewer due to the need for an improved dynamic grouping functionality. Furthermore, interviews with wind energy experts revealed which features are relevant for the community to describe wind farms. They can be divided into technical, temporal, terrain, and weather characteristics. From those we have covered, several insights were generated, including that the worldwide predominant land cover for the installation of wind infrastructure is agricultural land and that the predominant landform is flat terrain.
As digital technology continues to reshape communication and content consumption, collaboration and collective ideation have become fundamental to modern web technologies. Interactive articles aim to encourage active information exploration to foster knowledge creation. However, the integration of collaboration and tools like the infinite canvas remains unexplored in this context. This thesis investigates the opportunities and limitations of this area. It presents an exploration of the infinite canvas and examines its strengths and challenges for collaborative interactive articles. Through the design and implementation of an application built for Miro, a web-based platform utilizing the infinite canvas for creative collaboration, it investigates the design process for the infinite canvas. In a thinking-aloud user study, this thesis evaluates this application and its integration in this environment. This revealed that the infinite canvas offers an immersive virtual space, though collaborative artifact design introduces unique challenges. The distinction between shared and individual artifacts underscores the importance of considering the collaborative artifact space. Providing a clear user experience proved to be crucial, especially for inexperienced users. Tasks and input methods for end-users need to be kept simple. This thesis identifies two distinct user roles for the design of collaborative interactive articles: facilitators who guide group activities and participants who seek insights. It adopts the idea of Gamestorming as a framework for goal-driven collaboration to facilitate group work and provide engaging activities. Combined with semantic information present on an infinite canvas, this opens up further use cases. Moreover, information visualization techniques highlight possibilities to create engaging and interactive artifacts on the infinite canvas. In the context of visual analytics, the infinite canvas acts as a dynamic visual database, enabling direct interaction with data entities and relationships. Consequently, information visualizations can provide an overview of the infinite canvas and enhance the exploration of content. In the end, this thesis contributes to the collaborative and interactive potential of the infinite canvas, offering insights for the design of interactive articles and explorable visualizations.
In art historical research, the study of social networks can provide insights into complex interactions between artists. However, despite the successful use of network visualizations in the field of art history, there are still many open questions as well as challenges that need to be solved in order to support art historians in the best possible way for their research. For example, existing approaches offer only limited possibilities to visually explore art historical networks with regard to their geographical context as well as their temporal development. Furthermore, so-called node-link diagrams are usually used for the visual representation of networks, which in the past revealed weaknesses in comparison with other representations in various studies. Finally, it is difficult to evaluate applications with respect to their suitability in the art history domain, since established evaluation approaches are often not conducted with domain experts. We present Exhibitions Explorer, a solution which enables the interactive, visual exploration of art historical networks. The solution integrates different visualization approaches from research into a new visualization concept that focuses on the requirements of the domain. Using an insight-based evaluation approach, we also demonstrate that, provided a suitable visualization concept, network visualizations can lead to a high quantity and quality of domain-relevant insights.