Our work with the title Visual Comparison of Multi-label Classification Results concerns artificial intelligences that classify data items into different categories. The items can belong to multiple categories and, therefore, relationships are created between categories based on their mutual assignment. For each AI we visualize these relationships and compare them with the relationships of other AIs. This way we determine to which degree the classification of items overlap. We use application examples to demonstrate the different opinions of AIs with regard to detected food items in pictures and suggested genres for movies based on their posters.

The work was created in collaboration with the Luxembourg Institute of Science and Technology (LIST). It was presented at the VMV 2021 conference and the recorded talk is available on our youtube channel.

Investigating the implicit and explicit links that already exist in data-driven stories published by popular news media, the research revealed a set of best practices. These can help data journalists to design compelling and informative stories around certain data. Currently, we see many such articles published on the Covid-19 pandemic, which were also considered as examples in the conducted research [paper].

To further allow authors of interactive data-driven stories to easier link the text and diagrams, a web-based tool, Kori, was designed. It employs state-of-the-art natural language processing techniques to automatically detect the phrases that are related to parts of the accompanying diagrams. Besides, it offers an innovative user interface to link a textual phrase to visual marks in the charts [paper]. 

The research project applies these ideas in various application areas. Recently, a system to investigate the lives of historic figures was suggested where users can ask questions and the system replies using a mix of text and linked charts. While the answer is generated using an AI model (the GPT-3 language model), the system puts the answer in a context by visualizing data on a timeline and other views. This allows users to explore the data beyond just the information that was inquired [paper].

The research has been done in collaboration with national and international partners. It is published in Computer Graphics Forum (CGF) and IEEE Transactions on Visualization and Computer Graphics (TVCG), two of the leading visualization journals, as well as is presented at the EuroVis 2021 and IEEE VIS 2021 conferences. As part of the VGIscience lecture series, Shahid Latif summarized the research also in a single talk [video].

Clicking the image opens the video in a new browser window.

This work was done by Shahid Latif, Hagen Tarner, and Fabian Beck and will be published in an upcoming issue of IEEE Computer Graphics and Applications. Parts of this work originate in our Master Project Group Data-Driven Audio Guides for Virtual Reality.

More information, as well as the full paper can be found in our publication archive: Talking Realities: Audio Guides in Virtual Reality Visualizations.

Our tool visualizes the trains’ behavior on a timeline and shows connections between important regions of the network. The visualization tool was developed in collaboration with the Flatland community and with Günter Wallner, Assistant Professor at the Eindhoven University of Technology and Senior Scientist at the University of Applied Arts Vienna.

Check out the tool here and learn more about the Flatland award winners.

This work will be presented next week at the Celebrating the Scientific Value of Failure (FailFest) Workshop at IEEE VIS 2020. The work was done by PhD candidates Shivam Agarwal and Shahid Latif, and supervised by Prof. Dr. Fabian Beck. For more details, please check out our paper.

We propose a novel and general approach to visualize dynamically changing categories of data elements. Since elements can belong to more than one category, the categories overlap in various levels. We use a layered network for visualization, enriched with various visual marks showing all details of the data. The proposed approach is universally applicable to categorical data, for instance, can be used to analyze researcher’s field of interest, evolution of developer activities in a software project, changing market strategy of companies, etc.

The work was led by Shivam Agarwal and done in collaboration with colleagues from University of Stuttgart, Germany and Open University, UK. Project advisor Prof. Fabian Beck summarizes: “We have been working already for some time to make dynamic categorical data explorable. Now, this work, together with our also recently published approach, Set Streams, lays a foundation for a new way to visually analyze this kind of dynamic data.”

Hagen Tarner will present our approach on visualizing evolution and performance metrics at this year's Seminar Series on Advanced Techniques & Tools for Software Evolution (SATToSE). The document is available on our website.

The approach was developed in the DFG-funded project "SoftwareDynamics²" together with scientists from Austria. The prototype of the implementation was developed during the ViVaSD - Visualizing Variability in Software Dynamics project group in summer semester 2019, here at UDE.

The second visualization technique can show how an automatic method learns to get better and better in classifying image data into different categories. Changing classifications during this training process are depicted as visual streams that flow between the categories. The approach, called “Set Streams”, also supports other scenarios and applies to all sorts of overlapping categories that change across time.

Both approaches were recently presented at EuroVis 2020, and they will appear as two independent articles in the international journal Computer Graphics Forum. The conference was originally scheduled to be held in Norrköping, Sweden, but – like other conferences – had to transform into a virtual event, with interactive live video streaming sessions.