Developing Geo-temporal Context from Implicit Sources with Geovisual Analytics

Transcription

1 Developing Geo-temporal Context from Implicit Sources with Geovisual Analytics Brian Tomaszewski 1 1 The Pennsylvania State University, Department of Geography, 302 Walker Building University Park, PA, USA bmt139@psu.edu Abstract. The key Geovisual Analytics research question addressed in this paper is how knowledge of past situations can be computationally extracted from heterogeneous and implicit information spaces and presented as visual artifacts within an interactive environment to facilitate reasoning about situations from geographic and temporal (or geo-temporal) contextual perspectives. This paper in particular will examine the shared development of geo-temporal context from implicit geospatial and temporal references contained in open-source channels such as the news media to support situation assessment in crisis management activities. The paper presents the Context Discovery Application (CDA), which is a prototype collaborative, Geovisual Analytics situation assessment environment that facilitates the development of geographical and temporal context using implicit sources. Keywords: Geo-historical context, open-source information, geographic information retrieval, collaboration, crisis management 1 Introduction Geo-temporal context, as defined by this research, is information about the interconnectedness of phenomena, events, and place across multiple spatial and temporal scales within past situations that can provide meaning and background information to the understanding of present situations and insight into future situations. The importance of geo-temporal context is greater than ever. For application domains such as crisis management, geo-temporal context helps to provide insight and understanding to varied geographic reactions to a disaster event, develop post-event intelligence about what happened during a crisis and why, aid in collection of information about hazard mitigation discussions underway in various locals, and assess threats and vulnerabilities before disasters happen. Information sources that can potentially provide geo-temporal context to situations are vast and heterogeneous ranging anywhere from GIS layers, , text messages, camera-enabled cell phone pictures, online news reports, and beyond [18]. Therefore, the problem with developing geo-temporal context is not a lack of information, but rather, how relevant information is made available, presented, accepted, and understood by those who need it at the right time for the right reason. Geovisual Analytics offers a new scientific framework for developing an approach to ICA Commission on Visualization and Virtual Environments Annual Meeting; August 2 and 3, 2007; Helsinki, Finland

2 identifying relevant geo-temporal context information by using analytical process that combine human vision and cognition with computer-based visualization and computational tools and interfaces that can support flexible connections to relevant data and supporting knowledge, and are specifically designed to provide support for analytical reasoning [14]. 1.1 Motivation An area of inquiry that is particularly relevant for using a Geovisual Analytics approach is the development of geo-temporal context from implicit geospatial and temporal references in text documents. The notion of implicit geospatial and temporal references refers to largely qualitative, unstructured information contained in text documents such as textual/word references to locations of towns, counties or countries that can be combined with the largely numerical and structured information found in geospatial databases. Implicit geographical information contained in open-source channels such as the news media present an important source of geo-temporal context and situation information [11]. For application domains such as crisis management, implicit information sources such as the news media are used by organizations such as the Federal Emergency Management Agency (FEMA) and Reliefweb.org for geospatial intelligence analysis and situation monitoring/reporting [4, 5, 12]. Furthermore, developing geo-temporal context to support activities in application domains such as crisis management requires the application of analytic reasoning processes, such as the use of induction to understand the contextual dimensions of a situation based on the systematic reasoning with information sampled about that situation, which must operate within the inherent collaborative, team nature of work within these domains. It is with these motivations in mind that the Context Discovery Application (CDA) is being developed [17]. The CDA is a prototype situation assessment environment being developed with a long-term goal to facilitate the collaborative development of geographical and temporal context using implicit sources. The notion of the discovery of context implies that users will be able to find geographical and temporal context information that was previously unknown, and be able to share and synthesize this information with co-collaborators by applying combined expertise to finding relevant contextual information. 2 The Context Discovery Application 2.1 Features and Overview Current Geovisual Analytical functionality of the CDA includes automated retrieval of news stories based on a user-specified situation context, computational processing, geocoding and visualization of geographic place names and possible Page 2 of 12

3 relationships between places across user-defined geographic scales over time from within the stories retrieved. Furthermore, the CDA also supports formal ontology integration to find potentially relevant non-spatial dimensions within data retrieved. Geographic information presentation functionality includes tightly coupled map displays that allow users to simultaneously view geographical locations in a virtual globe and standard 2D cartographic perspectives for geographical orientation and analysis. Ongoing development of the CDA includes functionality to support realtime and asynchronous geocollaboration, shared document viewing, and temporal information interaction. 2.2 System Architecture The CDA uses a client-server environment and is implemented using Java, Google and Flex technologies. The client tier uses both an online interface and a desktop client in the form of Google Earth (GE). The server tier hosts the geospatial database used for place name queries, and the functionality for performing geoanalytical processing, OWL/ontology integration, geocollaboration, and construction of the output visualization that is rendered in the GE client. Figure 1 shows the general components of the CDA. Figure 1: CDA Architecture Overview Page 3 of 12

4 3 CDA Methods and Approach 3.1 Web Page Extraction Publicly available information outlets are used to gather information via Real Simple Syndication (RSS) feeds. A particular focus was made on using RSS feeds from Google News. Other information input types, such as flat files, can be used as well. Figure 2 shows the basic process of how the CDA extracts relevant textual content from a web page for processing. Figure 2: Extracting Relevant Web Page Content 3.2 Geographic Information Retrieval The CDA performs Geographic Information Retrieval (GIR) using a pattern matching algorithm and the Generic Architecture for Text Engineering (GATE) program [3]. GATE is used to perform textual annotation of named-entities (i.e locations, people, organizations) and parts-of-speech annotation (verbs, nouns) within the input Page 4 of 12

5 text. For locations in particular, GATE uses a gazetteer approach for annotating locations. As the default gazetteers of GATE do not provide effective geographic granularity for GIR at scales beyond large cities, the CDA dynamically adjusts GATE s location gazetteers based on a default sense of place for a search. For example, if news stories are being retrieved from Pennsylvania, the CDA will automatically update GATE s location gazetteers to include place names from Pennsylvania. The benefits of this approach are that irrelevant place names are not included for matching, thus potentially preventing false positives. The limits of this approach is that it requires there be a default or user provided sense of place for a document in order to adjust the gazetteers, or a sense of place that is reasonably bounded in scale (such as a state). Using the annotation provided by GATE, the CDA then attempts to construct geophrases that capture near-context word information around location words found, an approach similar to that discussed in [9]. The purpose of this step is to help with disambiguating locations found. For example, in the sentence: The flood will effect all of Springfield County. The word Springfield will be tagged as a location, but given the many possible Springfield s that exist (cities, towns, counties, businesses, etc.), it is difficult to determine which one to geo-code the word to. By identifying the term County after the explicit reference of a location, the geophrase Springfield County can be used to guide the geo-coding as the term County provides meta-information about the scale of the entity that was found. If a document has a default sense of place, this meta-information can help pinpoint a location within the extent of the default sense of place (i.e a state). If no default sense of place for a document exists or the default sense of place is very broad (i.e a country), this information is still of use, but the location will more difficult to determine and may need to be compared with other locations found in a document, such as a city that is potentially in the same state as the place that is to be geo-coded. 3.3 Google Earth Visualization Google Earth (GE) is used to present CDA search results in a virtual globe environment. The CDA creates custom views and renderings and integrates custom functionality through the Keyhole Markup Language (KML), an XML-based structure used to define customizations in GE. KML is used to render the results found by the CDA geocoding process in GE. Using a network link connection, the user can easily take the KML-based output renderings from the server for addition into GE. Figure 3 is a screen shot of a CDA rendering of news stories based on a query of West Nile Virus in Pennsylvania. Page 5 of 12

6 Figure 3: CDA Output rendered in Google Earth Each place found in the search is connected to the origin point using a line. In its current form, the origin point is based on an assessment made the CDA as to location of the entity publishing the news story. Future work will include assigning single or multiple origin points or footprints based on an assessment of locations found relative to themes found within a story [1, 13]. The thickness of the line indicates the number of times a place was referenced in the news story, point symbols represent the geographic scale of the entity found (town, county etc.). The transparency of the line indicates how old the story is relative to the time when the geovisualization was created. These approaches are used to give a quick overview of the information returned before removing unneeded information. 3.4 Maintaining Geographic Perspective The CDA provides overview and zoomed perspectives of geographic and other data being examined in order to maintain orientation and perspective by using a novel approach that links the views of GE with Google Maps (GM). GM in turn is contained within the Open Layers web mapping client software 1 (Figure 4). 1 Page 6 of 12

7 Figure 4: Linked 2D/3D maps views in the CDA GM provides a comprehensive, traditional 2D view rendered in a simple, yet visually pleasing cartographic style. When linked with the 3D/2D perspective of GE that uses satellite imagery as a backdrop, the two views provide a broader view of the geography being investigated. Furthermore, the scale between GE and GM is flexible where both can easily serve as either the overview or detail due to ease of scale manipulation. The CDA also supports the coordination of 2D/3D views with any other 2D map interface that can be viewed in the OpenLayers client. 3.5 Supporting Collaboration The CDA currently supports two forms of real-time and asynchronous geocollaboration. The intent of providing collaboration capabilities is to facilitate social interaction and group work among collaborators in order to develop geotemporal context. For example, by providing collaboration capabilities, the effectiveness of teams to collaborate in crisis-oriented information synthesis activities can be improved by applying combined expertise to finding relevant information and to interpreting results of searches. In particular, visual artifacts, are the means by which group work is facilitated in the CDA [6, 10]. Real-time geocollaboration in CDA includes a basic text-chat interface and the ability to collaboratively create maps. Collaborative map creation is facilitated through the sharing of Web Map Service (WMS) layers between clients (Figure 5). Page 7 of 12

8 Figure 5: Real-time collaborative map creation in the CDA In Figure 5, the user on the right (Suellen) requests a map layer from the user on the left (Brian). Using the CDA s collaborative map creation functions, Brian has a buffer layer drawn on Suellen s map. Note that both users have different base maps (Google Maps and Microsoft Virtual Earth), thus allowing both users to use their own base geographic data, yet allowing them to develop shared geographical perspectives and common ground on a situation through shared data layers. Current asynchronous collaboration capabilities currently include the capture, retrieval, and display of user map extents (Figure 6 and 7). Page 8 of 12

9 Figure 6: Asynchronous Review of User Map Extents in the CDA Figure 7: Visualizing Collaborator Map Extents in Google Earth (from [15]) Page 9 of 12

10 In Figure 7, a bounding box is used to track the map extents of collaborators examining news media reports of areas potentially affected by wildfires. Hue distinguishes between users and provides social awareness. Collaborator Alan (red) is examining the area around Okeechobe closely. Collaborator Cindy (yellow) can see Alan s area of interest, and can potentially change her area of interest, thus keeping their activities coordinated. The CDA also provides capabilities for users to share and view previous user web searches and documents collected from searches (Figure 8). This allows individual users to examine information items collected by others, and thus may not have been aware of, and to compare past documents with present documents for temporal perspective. Figure 8: Shared document interface in the CDA web client The overall intent of the CDA collaborative functionalities is to facilitate and mediate co-collaborator activity awareness, a dynamic process where information is shared and updated over time [2]. Activity awareness may in fact restructure the geotemporal context and overall social context being developed by a group [7]. For example, shared awareness of one collaborator looking at one area on a map or reviewing a particular piece of information may provide new insights into the geotemporal context being developed and modify activity behavior of other collaborators Page 10 of 12

11 such as looking at information sources previously not considered important or relevant. 4. Summary and Conclusion This paper has discussed how geo-temporal context can be collaboratively developed using a prototype Geovisual Analytic environment called the Context Discovery Application (CDA). The current focus of the CDA is to develop geo-temporal context from open-source information sources such as the news media. The CDA s Geographic Information Retrieval (GIR) and other computational procedures allows collaborators to mine open source information for locations of interest, and to examine, through visual representation and visual interfaces to information retrieved, possible connections between locations, and relations between locations and discrete temporal events and temporal concepts, thus showing, in part, the geo-temporal contextual dimensions for a situation of interest as reflected in open source information channels. Collaborative capabilities of the CDA facilitate the social process by which geo-temporal context is developed and shaped, for example allowing users to examine each others search results, collaboratively create maps to add additional geospatial context to situations, and to maintain awareness of cocollaborator activity in asynchronous time. Future research for the CDA will include the retrieval of documents and information that were once relevant within a particular past time, as opposed to the current procedures of understanding past events from present documents. An example of such an application might be retrieving daily situation reports about a given region from three years ago to understand a present situation. Other research directions will involve developing a formal model of geo-temporal context [16], and usability testing of the CDA with crisis management domain experts. Acknowledgments. The research reported here has been supported by the National Science Foundation under Grant EIA This work is also supported by the National Visualization and Analytics Center, a U.S. Department of Homeland Security program operated by the Pacific Northwest National Laboratory (PNNL). PNNL is a U.S. Department of Energy Office of Science laboratory. References 1. Amitay, E., Har El, N., Sivan, R. and Soffer, A.: Web-a-Where: Geotagging Web Content. In Proc. SIGIR 04. pp ACM, Sheffield, South Yorkshire, UK. (2004) 2. Carroll, J.M., Rosson, M.B., Convertino, G. and Ganoe, C.H.: Awareness and Teamwork in Computer-Supported Collaborations. Interacting with Computers, 18:21-46, (2006) 3. Cunningham, H.: GATE, a General Architecture for Text Engineering. Computers and the Humanities, 36(2): , (2002) 4. Department of Homeland Security: DHS Daily Open Source Infrastructure Report,< Accessed: 25 June (2007) Page 11 of 12

12 5. Federal Emergency Management Agency (FEMA): National Situation Update,< Accessed: 25 June (2007) 6. Garbis, C.: Exploring the openness of cognitive artifacts in cooperative process management. Cognition, Technology and Work, 4(1):9-21, (2002) 7. Greenberg, S.: Context as a dynamic construct. Human-Computer Interaction, 16(2, 3 & 4): , (2001) 8. Keim, D.A., Panse, C. and Sips, M.: Information Visualization: Scope, Techniques and Opportunities for Geovisualization. In: Dykes, J., MacEachren, A.andKraak, M.J. (eds.): Exploring Geovisualization. Elsevier (2004) Li, H., Srihari, R., Niu, C. and Li, W.: InfoXtract location normalization: a hybrid approach to geographic references in information extraction. Proceedings of the HLT-NAACL 2003 workshop on Analysis of geographic references-volume 1:39-44, (2003) 10. MacEachren, A.M.: Moving geovisualization toward support for group work. In: Dykes, J., MacEachren, A.andKraak, M.J. (eds.): Exploring Geovisualization. Elsevier (2005) Mubareka, S., Khudhairy, D.A., Bonn, F. and Aoun, S.: Standardising and mapping opensource information for crisis regions: the case of post-conflict Iraq. Disasters, 29(3): , (2005) 12. ReliefWeb: What We Do,< Accessed: 25 June (2007) 13. Silva, M.J., Martins, B., Chaves, M., Afonso, A.P. and Cardoso, N.: Adding geographic scopes to web resources. Computers, Environment and Urban Systems Geographic Information Retrieval (GIR), 30(4): , (2006) 14. Thomas, J.J. and Cook, K.A.: Illuminating the Path: The Research and Development Agenda for Visual Analytics. IEEE, Los Alametos, CA (2005) 15. Tomaszewski, B.: Cartographic and Visual Representation of Situational Information Created through Computational Extraction Procedures: Foundations for Awareness. In Proc. 23rd International Cartography Conference (ICC), Moscow, Russia (2007) 16. Tomaszewski, B.: Local Model Semantics, Categories, and External Representation: Towards a Model for Geo-historical Context. In Proc. Sixth International and Interdisciplinary Conference on Modeling and Using Context (CONTEXT'07), Roskilde University, Denmark (2007) 17. Tomaszewski, B.: Mapping Open-Source Information to Support Crisis Management. In Proc. First Annual DHS University Network Summit on Research and Education, Washington, D.C. (2007) 18. Tomaszewski, B.M., Robinson, A.C., Weaver, C., Stryker, M. and MacEachren, A.M.: Geovisual Analytics and Crisis Management. In: B. Van de Walle, Burghardt, P.andNieuwenhuis, C. (eds.): In Proc. 4th International Information Systems for Crisis Response and Management (ISCRAM) Conference, Delft, the Netherlands (2007) Page 12 of 12