KnoVA – Knowledge–Assisted Visual Analytics
The overall vision that drives this research proposal can be captured in the following working hypothesis/aim:
"The adoption of knowledge-assisted Visual Analytics methods, in which the different processes involved can rely on explicit prior knowledge, results in an improved data analysis, with a reduction of the cognitive load for users, and an increase in number and quality of their insights."
- Talin Barisani-Asenbauer (Laura Bassi Centres of Expertise LBC Ocuvac, Medical University of Vienna, Department of Specific Prophylaxis & Tropical Medicine, Center for Pathophysiology, Immunology & Infectiology)
Austrian Science Fund (FWF), grant P31419-N31.
Nowadays increasing amounts of data about complex phenomena are collected, stored, and made available for analysis. Visual Analytics (VA), “the science of analytical reasoning facilitated by interactive visual interfaces”, is a multidisciplinary approach to make sense of this data, combining the enormous processing power of computers with the outstanding perceptual and cognitive capabilities of humans. Users of VA systems need to rely on prior knowledge to gain insights from data, formulate and test hypotheses, interpret results, and discover new knowledge. Users’ tacit knowledge is taken into account for designing visualization methods, but the systematic utilization of explicit knowledge is largely unexplored. Existing knowledge-assisted approaches put a stronger emphasis on operational (how to interact) than domain (how to interpret) knowledge; they focus mainly on visualization and disregard other VA aspects. In the proposed project, we tackle the following research question: how can the VA process benefit from explicit knowledge in terms of enhanced interactive visualization and automated data analysis capabilities? Our hypothesis is that a knowledge-based VA approach, utilizing explicit knowledge across all VA stages, improves the analysis by increasing number and quality of the insights found by users while reducing their cognitive load.
The main scientific innovation of our approach lies on the enhancement of the VA process by making use of available knowledge repositories, also collected for different purposes than data analysis tasks. These knowledge bases, integrated within VA systems, will be exploited to assist not only visual mapping and interactive exploration, but also automated analytical methods. In particular, we will enhance VA methods by adapting the visualization and data mining processes in order to leverage available prior explicit knowledge and introducing complementary knowledge-assisted processes, like simulation and guidance.
We will apply human-centered design methods, in particular the well-established nested model for visualization design and validation. Despite the conceptual generality of our proposed approach, we will tailor it in specific application domains. In the medical domain, for example, we will use the knowledge formalized in computer-interpretable clinical practice guidelines to improve the analysis of electronic health records. A continuous involvement of domain experts will enable iterative refinements of developed artifacts (conceptual models, mock-ups, functional prototypes) and validation of results.