n a t h @ m i c r o s o f t . c o m

Participatory design session (c) Jean-Baptiste Labrune

y research focuses on the visualization and exploration of graphs and networks.

During my Ph.D. I investigated the potential of adjacency matrix-based representations (tables) to represent networks and attempted to merge them with the more traditional representations: node-link diagrams.

Recent interests include the visualization of heterogeneous networks (multiple types of nodes and edges), the visualization of networks evolving over time, and taking advantage of natural user interactions (sketch, natural language) to create and interact with visualizations.

Please, contact me for internships if you are interested in any of the research topics below.

Pursuing the work of Jacques Bertin

In my Ph.D. thesis, I designed alternative representations to node-link diagrams to represent graphs and networks. I investigated novel representations inspired by Jacques Bertin's reorderable matrices [Arts, Humanities, and Complex Networks Keynote]. Unlike node-link diagrams, adjacency matrices do not suffer from node overlapping or edge crossing issues.

With the increasing use of social networking sites such as Flickr or Facebook, rich and very large social networks are available for social scientists to analyze. Following a participatory design approach, I involved social scientists into the design of interactive visual systems based on adjacency matrices: MatrixExplorer [InfoVis 2006], MatLink [Interact 2007] and NodeTrix [InfoVis 2007]. I also developed interactive techniques to navigate in large matrices [CHI 2008] and visualize datasets of several millions of nodes such as Wikipedia [PacificVis 2008]. Finally, I applied these techniques to analyze 20 years of scientific collaboration in the field of HCI [IJHCI 2008].

Future directions include the use of matrix-based representations to explore different exciting application domains such as functional models of the human brain (fMRI data).


Visualization of social networks

Representing communities and groups

Visualizing groups and their elements is difficult when many elements belong to many groups. When analyzing social networks, it is often the case that many persons belong to many different communities. In these cases, representing groups with convex hulls (e.g. Euler or Venn diagrams) is often impossible.

We created a set of techniques to untangle Euler diagrams [InfoVis 2010]. In the first technique, we decompose groups in multiple regions and connect these regions using bundled links. Groups are nested within each other when they intersect. In the second approach, we avoid intersecting regions by duplicating elements in multiple groups.

Our most recent project investigated the use of a single continuous geometric line to represent a group instead of the traditional bubble. We discuss the design and report results from user studies showing that the technique is promising in [InfoVis 2011].


Evolution over time

Visualizing data evolving over time is exciting as the time-dimension is like no others. We initiated our research with an analytical tool for better understanding edits to Wikipedia pages with iChase [AVI 2010]. Then, interested in the dynamic aspect, we designed ResearchWave, an ambient visualization to show scientific collaborations to MSR researchers [DIS 2010].

We also played with simple visualizations such as tag clouds and envisionned how to show trends of their evolution over time in SparkClouds [InfoVis 2010]. We investigated how such simple dynamic visualizations could support learning in the classroom. In classsearch, we provide ambient evolving visualizations for helping students learn how to build and refine queries to search the web [CHI 2011].


Interacting with visualizations

Natural interaction techniques

To better understand how people naturally create and manipulate visual representations in their everyday life, we performed a field study analyzing about 80 whiteboards of researchers at MSR. Results and implications for designing visualizations are available here: [InfoVis 2011].

Building from this study, we are currently designing a pen and touch environment for creating and manipulating visualizations. Our preliminary system is based on sketch to create simple charts on whiteboard [ITS 2011].

Another on-going project seeks to understand how people describe visualizations in natural language. Learning from our observations, we aim at building a system allowing people to build visualizations by describing it textually or verbally using common language vocabulary.

Future directions include the design of tangible elements to create and manipulate visualizations.


Navigating in visualizations

When visualizing large visualizations such as matrices or node-link diagrams of thousands or millions of nodes, one needs to effectively navigate from one region of the representation to the other. Panning and zooming supports navigation but have drawbacks especially when visualizing graph data. Indeed, when exploring a graph, you often want to see two connected elements that may be very far away in the representation.

To solve this issue, we created Mélange [CHI 2008], a technique that allows users to fold the space between multiple focus regions. Folding the space provides contextual information about the intervening space allowing users to assess how far focus regions are for example.

We also created several techniques taking advantage of the topological information of graphs [CHI 2009]. Bring and Go brings adjacent nodes nearby when pointing to a node. Link sliding provides guided panning when continuously dragging along a link. Another project inspired from insets used in cartography [EuroVis 2011] utilizes the topological structure of the graph to draw dynamic visual insets for links leaving the edge of the screens. These insets show portions of the surrounding area of the targets.

Future directions include interaction techniques for multitouch surfaces.


Evaluating interactive visualizations

Evaluating interactive visualizations is a challenge as visual exploration systems aim at answering questions you did not even know you had about your data. To attempt to ease the evaluation of graph exploration systems, we developed a graph task taxonomy [BELIV 2006] and studied how to generate realistic graph datasets [www]. To implement logging a posteriori in java applications, we also investigated aspect programming and how to take advantage of this technology to retrofit an application[BELIV 2008].

We are currently investigating how simple methodologies used in psychology as well as physiological sensors could help researchers better assess the readability of visualizations. I wrote a position paper highlighting possible research directions [BELIV 2010].