On intelligible multimodal visual analysis

Abstract

Analyzing data becomes an important skill in a more and more digital world. Yet, many users are facing knowledge barriers preventing them to independently conduct their data analysis. To tear down some of these barriers, multimodal interaction for visual analysis has been proposed. Multimodal interaction through speech and touch enables not only experts, but also novice users to effortlessly interact with such kind of technology. However, current approaches do not take the user differences into account. In fact, whether visual analysis is intelligible ultimately depends on the user. In order to close this research gap, this dissertation explores how multimodal visual analysis can be personalized. To do so, it takes a holistic view. First, an intelligible task space of visual analysis tasks is defined by considering personalization potentials. This task space provides an initial basis for understanding how effective personalization in visual analysis can be approached. Second, empirical analyses on speech commands in visual analysis as well as used visualizations from scientific publications further reveal patterns and structures. These behavior-indicated findings help to better understand expectations towards multimodal visual analysis. Third, a technical prototype is designed considering the previous findings. Enriching the visual analysis by a persistent dialogue and a transparency of the underlying computations, conducted user studies show not only advantages, but address the relevance of considering the user’s characteristics. Finally, both communications channels – visualizations and dialogue – are personalized. Leveraging linguistic theory and reinforcement learning, the results highlight a positive effect of adjusting to the user. Especially when the user’s knowledge is exceeded, personalizations helps to improve the user experience. Overall, this dissertations confirms not only the importance of considering the user’s characteristics in multimodal visual analysis, but also provides insights on how an intelligible analysis can be achieved. By understanding the use of input modalities, a system can focus only on the user’s needs. By understanding preferences on the output modalities, the system can better adapt to the user. Combining both directions imporves user experience and contributes towards an intelligible multimodal visual analysis.