This document proposes dynamic insets for context-aware graph navigation. Dynamic insets display nodes just outside the visible screen area in an inset window based on their degree of interest, allowing users to see important surrounding context. An evaluation with 12 participants found dynamic insets significantly outperformed existing bringing-and-going techniques for tasks involving close and distant context in graphs. A follow up study with 6 participants tested dynamic insets on map and social network scenarios, finding they provide useful contextual navigation of large graphs.
8. Degree of Interest (DOI) Rank off-screen nodes Show N nodes with highest DOI in inset DOI functions Neighbors (DOI = 1) Neighbors divided by distance (DOI=1/d) GPS: nearby gas stations Airline: ticket price, travel time, #stops, etc Social network: graph metrics 8
20. Conclusions Dynamic insets provide context-aware graph navigation Insets show destination of edges leaving border of screen Design parameters include DOI functions, layout, occlusion Applications: map, networks, UML, etc User studies confirm usefulness 20
22. Thank You! Contact Information:SohaibGhaniSchool of Electrical & Computer EngineeringPurdue UniversityWest Lafayette, IN, USA E-mail: sghani@purdue.edu http://engineering.purdue.edu/pivot/ 22
Hinweis der Redaktion
So we ran the study and collected the time and correctness results.Correctness was uniformly high, more than 97%, so I will not go into more detail on this here.For time, we did see significant differences. This chart shows a summary of completion times where smaller values are better: put simply, users were much faster with DI (the yellow bar) than with BG (the blue bar).If we break this down by task type, the picture becomes a little clearer: Dynamic Insets are much faster than Bring and Go for all tasks except the count neighbors task, where BG is better.These are all significant differences.The fact that BG wins over DI is probably due to the fact that interaction is minimized for the count neighbors task in the BG case, whereas it does require some flipping and panning in the DI case.
The controlled study only looked at threespecific tasks in a very constrained setting, so we decided to perform a second user study just to look at usability and scalability aspects of the technique.