GeoJabber: Finding Significant Analytic Events in Collaborative Visual Analysis Sessions Frank Hardisty GeoVISTA Center, Department of Geography, The Pennsylvania State University ABSTRACT The integration of a peer-to-peer instant messaging system into visual analytic software allows automatic extraction of significant analytic events, such as inference drawing, causality determination, or hypothesis generation, during the course of an analysis. It does so by examining the textual communications between collaborators and marking those analytic events which are determined to be significant using term extraction and term matching. These events can be used as entry points into the analysis session, as a way to better understand both the subject of analysis (such as a possible Sarin gas attack), the collaborative behavior of the analysts, and patterns of tool use. This approach can potentially make visual analytics more productive through support for sharing fragments of reasoning among analysts. The GeoViz Toolkit introduced here is an open source software project that enables multivariate visual analysis of geospatial data. The open XMPP communication protocol (also known as Jabber) was used in the GeoViz Toolkit software to create a working prototype of a geocollaboration system, by creating extensions to Jabber to support tool state sharing, including geospatial aspects of tool state. Advantages and disadvantages of using XMPP vs. other implementation methods are detailed through a set of examples and discussion of those examples. KEYWORDS: Collaborative and Distributed Visualization, Human- Computer Interaction, Geographic Visualization, High- dimensional Data, Visualization system architectures, toolkits, and problem-solving environments 1 INTRODUCTION A recent report by the National Research Council asserts that “enabling collaborative work with geospatial information” is a key research challenge to support in activities in many domains, including homeland security [1]. A scenario that illustrates this would be that in the event of a terrorist attack, multiple people would need to coordinate their response to the attack at the same time. A map interface that supported input from multiple users could be critically important. The unfortunate truth is that current geospatial information technology does not support collaboration between users in different places working on the same problem at the same time. This paper presents a test system that would investigate how to mediate between users who are working on the same geospatial problem in different places. The core problem that this research attacks is “how can users’ actions in maps in different places be coordinated?” The issue is that if multiple users make coordinated changes to the map interface, for example the current extent being viewed, it is potentially disorienting to the map user. Imagine two users simultaneously trying to zoom in on different areas of a map. If the users’ maps are fully coordinated in the spatial extent that the maps show, the users might experience frustration and conflict. Two alternative options for mediating the conflict are having persistent settings and to have a selectable history of map extents. The persistent setting approach is to have “Leader” and “Follower” settings on each map. If one user sets his or her map to the “Leader” setting, and the other user sets his or her map to the “Follower” setting, the leaders’ map extent would automatically be reflected in the followers map. The alternative, selectable history, method would be to give each user a clickable list of spatial extents, and allow the user to click on them to apply them. These approaches have shown promise in other problem domains [2]. In the remainder of this paper, first the GeoViz Toolkit is introduced, next a typology of what is to be coordinated between users during a collaborative analysis session is introduced, then the GeoJabber implementation is presented, with other implementation alternatives described. 2 THE GEOVIZ TOOLKIT The GeoViz Toolkit is an Open Source project which enables multivariate exploration of geospatially referenced data sets. One configuration of its user interface can been seen in Figure 1 below. Figure 1: GeoViz Toolkit Interface 302 Walker Building, University Park, PA 16802 hardisty@psu.edu