Publikationen von Jan-Frederik Kassel

A visualization recommender supports the user through automatic visualization generation. While previous contributions primarily concentrated on integrating visualization design knowledge either explicitly or implicitly, they mostly do not consider the user's individual preferences. In order to close this gap we explore online learning of visualization preferences through dueling bandits. Additionally, we consider this challenge from a usability perspective. Through a user study (N = 15), we empirically evaluate not only the bandit's performance in terms of both effectively learning preferences and properly predicting visualizations (satisfaction regarding the last prediction: μ = 85%), but also the participants' effort with respect to the learning procedure (e.g., NASA-TLX = 24:26). While our findings affirm the applicability of dueling bandits, they further provide insights on both the needed training time in order to achieve a usability-aligned procedure and the generalizability of the learned preferences. Finally, we point out a potential integration into a recommender system.

Conversational interfaces (CIs) have the potential to empower a broader spectrum of users to independently conduct visual analysis. Yet, recent approaches do not fully consider the user's characteristics. In particular, the objective of matching the user's language has been understudied in visual analysis. In order to close this gap, we introduce an answer space motivated by Grice's cooperative principle for framing personalized communication in complex data situations. We conducted both an online survey (N=76) to analyze communication preferences and a qualitative experiment (N=10) to investigate personalized conversations with an existing CI. In order to match the user's language properly, our results suggest to consider additional user characteristics along with their knowledge level. While mismatching communication preferences triggers negative reactions, a preference-aligned communication evokes positive reactions. As our analysis confirms the importance of matching the user's language in visual analysis, we provide design implications for future CIs.

Modern technologies enable data analysis in scenarios where keyboard and mouse are not available. Research on multimodality in visual analytics is facing this challenge. But existing approaches consider exclusively static environments with large screens. Therefore, we envision Valletto, a prototypical tablet app which allows the user to generate and specify visualizations through a speech-based conversational interface, through multitouch gestures, and through a conventional GUI interface. We conducted an initial expert evaluation to gain information on the modality function mapping and for the integration of different modalities. Our aim is to discuss design and interaction considerations in a mobile context which fits the user's daily life.

Exploratory data analysis is an essential step in discovering patterns and relationships in data. However, the exploration may start without a clear conception about what attributes to pick or what visualizations to choose in order to develop an understanding of the data. In this work we aim to support the exploration process by automatically choosing attributes according to an information-theoretic measure and by providing a simple means of navigation through the space of visualizations. The system suggests data attributes to be visualized and the visualization's type and appearance. The user intuitively modifies these suggestions by performing swiping gestures on a tablet device. Attribute suggestions are based on the mutual information between multiple random variables (MMI). The results of a preliminary user study (N = 12 participants) show the applicability of MMI for guided exploratory data analysis and confirm the system's general usability (SUS score: 74).

The amount of logistical data in the automotive industry drastically increases due to digitalization and data that is automatically generated due to Auto-ID-Technologies. However, new methods need to be devised to make sense of this data, in particular when users are mobile, and when users need to collaborate to solve complex logistical tasks, such as resource scheduling. We propose a visualization method for hierarchical event data that is designed for tablets. The main design goals have been to foster collaboration and enable mobility. Our think aloud user study shows that both the event recognition and understanding of the participants improved with the proposed solution.