Towards Interactive Visualization of Longitudinal Data to Support Knowledge Discovery on Multi-touch Tablet Computers

Andreas Holzinger, Michael Schwarz, Bernhard Ofner, Fleur Jeanquartier, André Calero Valdez, Carsten Röcker, Martina ZiefleCD-ARES 2014 · DOI

TL;DR

This paper explores how multi-touch tablet computers can revolutionize the interactive visualization of longitudinal data, particularly in data-centric fields like medicine. By leveraging the intuitive and rich interaction capabilities of touchscreens, the authors propose methods to enhance knowledge discovery from complex, time-dependent datasets, offering significant advantages over traditional mouse-based interfaces.

What Problem Does This Paper Address?

The paper tackles the growing challenge of managing and extracting insights from the ever-increasing volume of longitudinal data, especially in data-centric medicine. This type of data, recorded over time, is often complex, heterogeneous, and weakly structured. Traditional data handling methods struggle to keep pace with this data flood, leading to potential inaccuracies and user dissatisfaction. The authors highlight the need for efficient, user-friendly solutions that empower domain experts to discover knowledge effectively. They specifically focus on the potential of multi-touch tablet devices as an interaction medium, arguing that their unique capabilities offer distinct advantages for visualizing and interacting with longitudinal data compared to standard desktop interfaces.

Methods

The paper discusses various established visualization techniques for longitudinal data, including:

  • Sequence Graphs: Representing time-dependent data on a single dimension, useful for identifying event patterns.
  • Point Graphs: Extending sequence graphs by adding a second axis to represent another data dimension, allowing for visualization of two-dimensional data.
  • Bar Graphs: Replacing points with bars to improve comparability between data points.
  • Line Graphs: Connecting data points with lines to emphasize temporal trends.
  • Circle Graphs: Mapping time series data into a spherical domain, ideal for visualizing periodic data.

Beyond these visualization types, the paper emphasizes the crucial role of interaction methods for effective knowledge discovery. They discuss standard interaction techniques such as:

  • Zooming: Allowing users to get a high-level overview and then drill down into specific subareas.
  • Scrolling: Navigating through data when the display area is limited.
  • Focusing and Linking: Providing zoomed-in views with enhanced visualization techniques.
  • Brushing: Displaying pop-ups with additional information on rollover.
  • Filtering: Removing irrelevant data objects.

A key contribution is the exploration of how multi-touch gestures can implement these interaction methods more intuitively and efficiently than traditional mouse-based interfaces. They highlight gestures like pinch-to-zoom, double-tap for discrete zooming, one-finger drag for scrolling, and a "cut gesture" for selecting specific graph portions. The paper argues that the increased degrees of freedom offered by multi-touch input allow for the encapsulation of complex graph interactions into simple, learnable gestures.

Datasets

The paper does not introduce or evaluate on specific datasets. Instead, it focuses on the general principles and techniques for visualizing and interacting with longitudinal data, using illustrative examples for each visualization type.

Key Insights and Ideas

  • Multi-touch tablets offer a paradigm shift for longitudinal data visualization: The intuitive nature of touch input, combined with gestures, allows for more direct and fluid interaction with complex time-series data than traditional mouse-based interfaces.
  • Richness of interaction is key for knowledge discovery: Simply displaying data is insufficient; users need powerful tools to explore, manipulate, and understand it. Multi-touch devices excel at providing this rich interaction.
  • Gestural interfaces can simplify complex interactions: By mapping common visualization tasks (like zooming, scrolling, and focusing) to intuitive gestures, users can learn and apply these interactions more quickly, reducing cognitive load.
  • Longitudinal data visualization is particularly challenging in domains like medicine: The inherent complexity and time-dependent nature of medical data make effective visualization and interaction tools critical for diagnosis, treatment planning, and research.
  • Bridging the gap between mathematical analysis and visual inspection is vital: The paper suggests that future interfaces should support users in leveraging visual information to adjust model parameters, integrating analytical and visual approaches.

Key Results

The paper does not present quantitative results from experiments or evaluations. Instead, it offers a qualitative exploration of visualization techniques and interaction methods. The "results" are primarily conceptual:

  • A comprehensive overview of common longitudinal data visualization types and their use cases.
  • A detailed discussion of interaction methods that enhance data exploration.
  • A strong argument for the superiority of multi-touch interfaces for implementing these interaction methods due to their richer input capabilities and gestural commandos.
  • Identification of several "open problems" that require further research.

Limitations and Future Work

The authors acknowledge several limitations and identify promising avenues for future research:

  • Selecting appropriate axis parameters: This is identified as a "grand major challenge" when dealing with interactive visualizations of longitudinal data.
  • Handling missing data: Uncertainty in datasets, especially in the medical domain, is a crucial issue that needs robust solutions.
  • Scalability and platform issues: As data volumes grow, solutions for scalability and platform independence are necessary, particularly for mobile devices.
  • High-dimensional longitudinal data: Visualizing changes over time, trends, predictions, and correlations in high-dimensional data remains an open problem.
  • Optimal data resolution: Finding methods to choose optimal data resolutions (sampling frequencies) for both recording and visualization, based on anticipated data variability and pattern cycle lengths, is an ongoing challenge.
  • Dynamic visualization resolutions: Developing methods to handle changing visualization resolutions during continuous zooming, especially for merging multiple data points, is needed.
  • Integrating mathematical analysis and visual inspection: Creating user interfaces that allow users to utilize visual information for adjusting model parameters is a key future direction.
  • Leveraging other sensor inputs: Incorporating sensors like gyroscopes and accelerometers from multi-touch devices to create more state-dependent and context-aware interfaces.

Summary updated

BibTeX
@inproceedings{holzinger2014towards,
  author = {Andreas Holzinger and Michael Schwarz and Bernhard Ofner and Fleur Jeanquartier and André Calero Valdez and Carsten Röcker and Martina Ziefle},
  title = {{Towards Interactive Visualization of Longitudinal Data to Support Knowledge Discovery on Multi-touch Tablet Computers}},
  booktitle = {CD-ARES},
  year = {2014},
  doi = {10.1007/978-3-319-10975-6\_9},
}
Figure