This paper presents the 'Media Watch on Climate Change', an interactive Web portal that combines a portfolio of semantic services with a visual interface based on tightly coupled views. The interface enables users to access a repository of environmental knowledge, built by crawling about 300,000 news media articles in weekly intervals. Tagging services annotate each article to create a contextualized information space. A system component for automatically extending ontologies helps structure the knowledge repository, and feeds an ontology-based visualization module. The Web portal is intended to facilitate the access of complex datasets, help users navigate large collections of Web documents based on similarity, geospatial context and underlying concept, and provide annotations via standardized interfaces for third-party applications.
The widespread acceptance of three-dimensional virtual globes such as Google Earth and NASA World Wind advances the vision of a Geospatial Web. The following hypothetical scenario outlines how an integration of geospatial and semantic data may radically change working environments, impact workflows within and across organizations, and enrich individual interactions.
“Kathryn O’Reilly is a freelance editor who sells her ability to gather, filter and prioritize electronic content. In a virtual world built on contextualized information spaces, Kathryn seamlessly switches between geographic and semantic topologies. She begins her typical working day floating in the virtual space above Earth, ready to navigate the globe and semantic structures via subtle movements of her eyes. From her elevated position, Kathryn not only observes the rise and decay of topics, but also the unfolding of social structures based on the unique social networks of her friends and contacts. Across these networks she builds and shares her knowledge repository, and composes media products that are continuously being validated and enriched by the latest news feeds and third-party sources” [1, p.3f].
This scenario exemplifies the profound impact of the Geospatial Web on managing individual and organizational knowledge. The Geospatial Web will not only reveal the context and geographic distribution of location-based resources and services, but also catalyze virtual communities by matching people of similar interests, browsing behavior, or geographic location. 2
System Description
The vision of a Geospatial Web promotes the convergence of geographic information,
Internet technology and social change. Taking a step towards this vision, the Media
Watch on Climate Change uses automated content analysis to extract geospatial context
and build a geotagged knowledge base. The interface provides various means to
interactively access this knowledge base. It shows that geobrowsers are not only suited to
explore geographic features, but can also render other types of imagery such as
twodimensional semantic maps or three-dimensional knowledge planets. Such maps are
representations of semantic information spaces based on a landscape metaphor [
To increase awareness and the availability of environmental information, the IDIOM Media Watch on Climate Change provides a comprehensive and continuously updated account of media coverage on climate change and related issues. The portal aggregates, filters and visualizes environmental Web content from 150 Anglo-American news media sites. It is available at the following address:
http://www.ecoresearch.net/iswc2007 2.1
The knowledge base underlying the Media Watch on Climate Change is a contextualized information space. If authors and information providers do not provide contextual attributes, annotation services attempt to tag and classify new objects. If contextual attributes are provided, the annotation services validate the manual entries, add missing information, and suggest changes if conflicting information is found. The annotation services of the current prototype provide metadata along three dimensions: (i) spatial distinguishing between source and target geography; (ii) semantic - assigning the most relevant concepts from a controlled vocabulary; and (iii) temporal - adding timestamps for the reported event, the initial publication and subsequent revisions. Each element of the contextualized information space can be organized, indexed, searched and navigated along these dimensions.
Figure 1 shows a screenshot of the current prototype. The content view in the upper left window contains the active document, including its mirror date and source/target geography. Below, just-in-time information retrieval agents list documents referring to similar topics and nearby locations. Clicking on the related references extends the quoted text, clicking on the circular marker on the left activates that particular document. The semantic and geographic maps facilitate access to the underlying knowledge base. The information landscape in the upper right window shows the semantic associations between documents. Peaks indicate clusters of documents about a popular topic, whereas valleys represent sparsely populated parts of the information space. The two links in the upper left corner of this window allow users to switch between the semantic map and a visual representation of an automatically generated domain ontology (see Section 2.5). Below, a geographic map shows the locations referred to in the listed documents. Clicking on the ’max’ links increases the size of the maps.
Each of the ’3D’ links opens a new window. The semantic map links to a description of knowledge planets as a new interface metaphor. The link of the geographic map points towards a KML (Keyhole Markup Language) file that requires a geobrowser to display the 100 most relevant news items on a three-dimensional virtual globe.
Once users enter a search query, an additional window displays quotes including the target term as ’concordances’ (centering the target term and showing its immediate context in the various documents). Besides entering query terms, users can click on any position in the maps (not only on the document markers) to retrieve articles related to that particular location, topic, or domain concept. The different views are therefore said to be ’tightly coupled’: User actions in one window trigger an immediate update of all other displays.
Microformats like hCard, hCalendar, XFN and others provide straightforward techniques to embed semantically rich information into existing Web pages. Their markup is not as advanced and powerful as RDF, but aggregator services like Technorati have demonstrated how useful these tagging formats content have become. Within fifteen months, from November 2005 to April 2006, the number of tagged postings grew from zero to 100 million5.
Web2.0 sites like Dopplr.com6 or CouchSurfing.com7 started adapting microformats for building social networks using XFN and the hCard standard. “IBM is another early adaptor, enriching employee listings with hCards information. Unfolding the power of a decentralized and easy tagging architecture, Web developers and domain experts specify new domain-specific microformats in order to provide users with a steadily growing vocabulary for annotating Web content. Efforts like the W3C’s development of RDFa try to combine the advantages of micro-formats with the expressiveness of RDF.
’Geo’ microformats describe the target geography of the IDIOM Media Watch’s documents by encoding geographical information. ’Rel-tag’ markup extends existing hyperlinks, marking the destination as a system-designated tag. Data from the just-intime information retrieval agents is annotated using the ’hReview’ microformat, marking it as a review and thus giving it additional information such as a relevance rating and a review date. Browser add-ons such as the “Operator” and “Tails Export” Firefox extensions provide a user-friendly interface for importing such entries into existing applications. Geo-metadata from the IDIOM Media Watch prototype can be presented as a link to mapping applications such as Yahoo! or Google Maps, ’rel-tag’ markup refers the user to community sites such as del.icio.us8 and flickr9. RSS feeds provide annotated document feeds to topics and keywords promoting interoperability with third party applications. The feeds are available at the following address: 2.4
Generated by orthographically projecting and tiling semantic maps (= visual representations of semantic information spaces based on a landscape metaphor), knowledge planets allow visualizing massive amounts of textual data. At the time of map generation, the planet’s topology is determined by the content of the knowledge base. The peaks of the virtual landscape indicate abundant coverage on a particular topic, whereas valleys represent sparsely populated parts of the information space.
VisIslands, a thematic mapping algorithm similar to SPIRE’s Themescape [
The need for controlled vocabularies and shared meaning suggests that ontologies
shared conceptualizations within a specific domain [
Domain Expert(s)
Hearst Patterns
Sentence-Level Page-Level
WordNet Seed Ontology Extended Ontology
Spr/Activation
Spr/Activation Head Nouns
WordNet
Due to the sheer volume of data, the use of ontologies in large information spaces is difficult without an ontology extension system that suggests concepts and relations automatically. Figure 2 presents a conceptual view of the ontology extension architecture underlying the Media Watch on Climate Change.
A small set of terms obtained from domain experts or known ontology repositories
is first selected as a seed ontology. The seed ontology terms are then fed into the
analyzer, which is distributing the input to different plugins that provide evidence sources,
suggesting promising new concepts. Currently three evidence sources are considered:
1. Co-occurrence analysis [
Ontologies retrieved from the ontology extension process are serialized in OWL and then fed into the visualization framework. Transforming relations annotated using reification and all other information encoded in the serialized version of the ontology into the graphviz10 description language yields an SVG-file with a graphical representation of the ontology. The framework extracts the concepts’ positions from the SVG files, which provides the mapping required for interactions with the ontology view.
Applying the iterative ontology extension process to 50 Web sites mirrored in July 2007 and comprising approximately 80,000 documents yielded the Web portal’s domain ontology (see Figure 3). Arrows indicate confirmed hierarchical relations. The dotted lines connect semantically related terms whose exact type of relation could not be determined automatically. For these non-hierarchical relations, the (r) values indicate their strength based on the link assignment’s spreading activation level. High values suggest a strong relation between the concepts, a value of 8 being the maximum due to the specific setup of the spreading activation network.
The current navigational aid contains unlabeled associations, only indicating the strength of the relation but not its specific type. The IDIOM research project is currently extending the underlying algorithm to detect a much broader set of relation types 10 http://www.graphviz.org/ Current applications only hint at the true potential of geospatial technology to build and maintain virtual communities and to revolutionize the production, distribution and consumption of media products. Automatically annotating content from heterogeneous sources creates knowledge repositories spanning multiple dimensions such as space, time, and semantics.
Virtual globes improve the accessibility and transparency of such complex
repositories. Their introduction has popularized the process of “annotating the Planet” [
The Media Watch on Climate Change demonstrates how geospatial interfaces can be used to visualize layered representations of thematic maps and domain ontologies as a frontend for semantic services.
The Media Watch on Climate Change has been developed as part of IDIOM (Information Diffusion across Interactive Online Media; www.idiom.at), a research project funded by the Austrian Ministry of Transport, Innovation & Technology (BMVIT) and the Austrian Research Promotion Agency (FFG) within the strategic objective FIT-IT (www.fit-it.at).