Visualization Literacy in Early Education


During my post-doctoral studies at Koc University, Turkey, I started to work on visualization tools for tablet equipped classrooms. In collaboration with other researchers from Microsoft Research and INRIA, I worked on designing interactive interfaces to improve visualization literacy of children in early education. We employed field studies and focus groups at initial stages of this project, to understand the role of visualizations in early education, and to identify the key challenges faced when teaching visualizations at this level. We followed up with a survey study and quantitative analysis of existing visual materials used in the textbooks. This formative study revealed opportunities for improving visualization literacy at schools through tools that fuses well-known interaction techniques in visualization with existing pedagogical strategies to teach abstract concepts in general. Following an iterative design process I developed a web application not only enhances teaching of abstract visualizations, but also answers the teachers' need of quickly customize interactive material. The research paper was awarded an honorable mention at CHI 2017. More details of the project can be found here: https://cestlavis.github.io/.


Enriching Abstract Visualizations with Visual References

 

 

A node-link diagram enriched with contour structure generated according to the graph layout. 

A node-link diagram enriched with contour structure generated according to the graph layout. 

Exploring large visualizations that do not fit in the screen raises orientation and navigation challenges. Structuring the space with additional visual references such as grids or contour lines provide spatial landmarks that may help viewers form a mental model of the space. However, previous studies report mixed results regarding their utility. While some evidence showed that grid and other visual embellishments improve memorability, experiments with contour lines suggest otherwise. 

 

In this work, we describe an evaluation framework to capture the impact of introducing visual references in node-link diagrams. We present the results of three controlled experiments that deepen our understanding on enriching large visualization spaces with visual structures. 

Our findings shed a new light on previous results in the literature. In particular, our results seem to indicate that when node-link diagrams present recognizable motifs, such as clusters, the benefits of grids for revisitation tasks may not be discernible. Our studies are also the first ones to reveal tangible benefits of the use of contour lines to enrich node-link diagrams. Our experimental design attempts to capture how these visuals could help the viewer form a mental model of the visualization (by structuring the space and identifying spatial landmarks). Results indicate that contour lines play a significant role in these activities.

 

Brain Connectivity Visualization and Weighted Graph Comparison Techniques

As an intern at AVIZ Inria, I started to work on brain connectivity visualization. Brain connectivity consists of simultaneous activation associations between specific parts of the brain and neuronal fibers running across various points in the brain. Although a structural/spatial data set, brain connectivity can be expressed as weighted graphs. Scientists use abstract (non-spatial) node-link and matrix representations to be able to identify patterns in the connectivity data through comparisons. In this work we looked at how different graph representations could be augmented to support comparison of weighted graphs. 

More specifically, we identify common visual analysis tasks that neuroscientists carry out in brain connectivity analysis based on interviews with neuroscientists and an in-depth review of the domain literature. By doing so, we provide a link from domain-specific problems in neuroscience to more generic problems in HCI and visualization. Based on this task identification, we also establish that weighted graph comparisons can benefit a group of higher-level tasks in visual brain connectivity analysis. We explored alternative visual encodings that facilitate the comparison of edge weights across two graphs in a superimposed view, both in node-link diagrams and adjacency matrices.

The results of this study appeared at CHI in 2013, winning a Best Paper Award. I certainly believe many scientific data sets can benefit from looking at them in an abstract information visualization perspective. The challenge of how to associate the abstract representation with the structural/spatial representation remains as a topic that I would love to pursue in the future. 

 

 

Visualizing Sets

I worked on visualization of categories of elements which can be expressed as intersecting sets. Traditionally sets are represented with enclosing geometries. However, when a large number of sets intersect, the resulting set shapes and overlaps can become quite complex, and interpreting such geometries with occlusion becomes challenging. To limit the visual clutter and increase the readability of complex set representations, LineSets use geometrically continuous lines visiting all members of a set. 

LineSets minimize the clutter of intersecting sets by producing line crossings instead of geometry overlaps.

To better understand the advantages and
drawbacks of this concept, we performed a quantitative user study assessing its readability compared to state of the art, Bubble Sets by Collins et al.

Results show that LineSets help users for certain tasks where traditional set representations do not scale well. We present two applications of LineSets in the context of search tasks on maps and community analysis in social networks.

Later I collaborated with the authors of Kelp Diagrams.  The resulting Kelp Fusion method has significant advantages over both Bubble Sets  and Line Sets methods. 

 

 

Stereoscopic Highlighting

3D information visualizations often fail due to problems related to perspective scaling and occlusion. However, it is unfortunate that information visualization applications do not take advantage of the graphics technology improving every day. One of the main questions that drive my PhD work was to identify niche problems where graphics technology could enhance conventional information visualization techniques. Stereoscopic highlighting was a result of this quest.  

Stereoscopic highlighting is a technique that utilizes stereoscopic depth to highlight regions of interest in a 2D graph by projecting these parts onto a plane closer to the viewpoint of the user. The technique aims to isolate and magnify specific portions of the graph on demand without resorting to other highlighting techniques like color or motion. This mechanism of stereoscopic highlighting also enables focus+context views by juxtaposing a detailed image of a region of interest with the overall graph, which is visualized at a further depth with correspondingly less detail.

 

Text Visualization on Large Displays

 

A large display (on the right) is coupled with individual tablets (on the left) for enhanced text viewing. 

A large display (on the right) is coupled with individual tablets (on the left) for enhanced text viewing. 

Text on large displays (or rather on a tablet): Many data sets contain structural 3D components along with associated textual data. 

While structural data is often best visualized on large stereoscopic displays, presenting textual information on large displays poses problems. Main issue is a trade off between size dependent legibility of text and the occlusion of structural data if text is presented at larger sizes. Our solution to this problem integrates structural data shown on a large display while users select features of the structure and view the associated textual data on personal tablets.

 

 

Text Summary

Visualizing a summary of reviews on a retail website for a single product. 

Visualizing a summary of reviews on a retail website for a single product. 

Every day, vast numbers of customer reviews are accessed on retail or online services websites. Although valuable information can be gleaned from reviews, these sources are underutilized both by consumers and businesses due to their unstructured nature, serial presentation, limited search tools, and low ratio of useful information to the overall amount of data (high noise). 

OpinionBlocks is an interactive visualization tool to help people better understand customer reviews. It is designed to progressively disclose increasingly detailed textual information from various reviews while continuously providing visual graphical summaries. The visualization initially exposes text at the keyword level, and then exposes snippets that the keywords are used in, and finally shows the snippets within the context of an entire review. We provide interactive tools to let users navigate across different text granularities in an intuitive way and additionally, utilize smooth animations to preserve context across views.

 

Allosphere

 
 

During my PhD, I was involved with the Allosphere project. We collaborated with many different groups of scientists, creating immersive visual representations of their data sets within the Allosphere