A NEW TOOL FOR GIS
FREEDOM OF INFORMATION: MULTIVALENT DOCUMENTS
PEER REVIEW THROUGH COLLABORATIVE FILTERING
he traditional system for handling scholarly information is on a collision course with the exponential growth of this information. Even spending 15% more each year, university libraries are simply unable to keep up as this information doubles about every four years, overwhelming both the technical and financial models that have worked until now. Both to overcome these "growing pains" and to explore the promise of fundamental improvements that modern information technologies might provide, UC Berkeley researcher Robert Wilensky and colleagues are rethinking the entire life cycle of scholarly information, from creation to collaboration, review, dissemination, and archiving, as part of the NSF Digital Libraries Initiative, Phase 2 (DLI2).
Information technologies have had major impacts on scholarly communication in areas from research and publishing to libraries. However, many applications have involved improvements in traditional ways of doing things rather than qualitatively new approaches. Is the academic community overlooking important new possibilities in digital technologies that could radically enhance the scholarly information life cycle? Wilensky and colleagues in the UC Berkeley Digital Library project think so, and they are pursuing research on a number of fronts that shows promise of fundamentally transforming scholarly communication.
"We envision wide applications of new information paradigms across many areas of society, but initially we're focusing on scholarly communication, both because of the need in light of steeply rising costs, and because there are fewer barriers to change since academic researchers are naturally interested in wide circulation of the knowledge they create, with fewer impediments like copyright concerns," said Robert Wilensky, principal investigator of the UC Berkeley Digital Library project and Professor, Computer Science Division and School ofInformation Management and Systems.
As part of DLI2, the UC Berkeley Digital Library project is addressing an array of issues in scholarly communication involving both software and data collections, some in the InterLib collaboration between UC Berkeley, UC Santa Barbara, and Stanford University that is demonstrating technologies in the California Digital Library (CDL) as well as in a testbed developed by SDSC.
Top | Contents | Next
In developing a general document model, Wilensky and colleagues wanted to include information types beyond text, enabling researchers to seamlessly federate and interact with any kind of information. "Because most people don't consider geographic data to be documents-things like maps or spatial data or digital raster graphics or rectified images-working with Geographic Information System (GIS) data allowed us to broaden the definition of what people think of as documents," said Wilensky.
In addition to being able to access a wide variety of data formats, Wilensky and researchers including Loretta Willis, Jeff Anderson-Lee, and Howard Foster of the UC Berkeley Digital Library project developed GIS Viewer 3.0 as a Java applet that supports a number of powerful capabilities, including displaying and turning on and off multiple layers of geopositioned data; displaying vector, raster, and point data; smooth zooming and panning; hyperlinking points and regions; interacting with other online systems through server-side, Common Gateway Interface (CGI) scripts; and finally, distributed end-user annotation and saving. Figure 1 shows the application's wide zoom capability.
The project has evolved into a substantial tool that researchers are beginning to use in numerous applications. "For example, the GIS Viewer interoperates with all the data on the TerraServer, which contains the U.S. Geological Survey aerial photos and high resolution graphic map images for the entire United States. That's a terabyte-sized data set, and we're pleased that the GIS Viewer is a serious tool able to work at that scale," said Wilensky.
While these capabilities are interesting in themselves, the GIS Viewer is a special case of a more general and powerful approach the researchers have been working on for several years, including what they call multivalent documents, which enable completely new modes of scholarly communication.
Top | Contents | Next
Multivalent Document "Lens" Function for Scanned Text
The multivalent document model enables seamless access to different data types, including scanned text. Shown is the "lens" function, capable of applying optical character recognition and magnification to user-selected areas. In a scanned page "enlivened" as a multivalent document, it is also possible to select and paste text, highlight matching search terms, and perform other manipulations such as sorting a table.
Imagine that it is possible to seamlessly open and work with HTML, PDF, or any word processing file, as well as scanned or legacy documents. Imagine it is possible to compose these different document types, overlaying them with data retrieved from distant databases. Imagine it's possible to "semantically align" these documents, as well as alternately hide, display, or modify them in different ways the user selects. Imagine it's also possible to perform complex operations including sophisticated searches, analyses, and visualization options, along with more traditional word processing and editing operations. Then, imagine that these documents can be annotated, in a new document, as easily and flexibly as writing notes on a paper page. And if the software lacks a needed capability, it can be added by those who need it.
To achieve these capabilities, the researchers have used a middleware model with structured document layers and functions, providing an "anytime, anywhere, any type, every way user-improvable digital document platform." This approach provides freedom from traditional barriers of different file formats and data types. In addition to reducing fragmentation in data, the model is also designed to help overcome functional fragmentation or limitations in applications. Traditionally, new functions can be added only by the developer, and are neither consistently implemented nor uniformly available across different applications. The approach Wilensky and colleagues have taken is to invert traditional application design, specifying only a minimum set of protocols, with all major functions moved into extensions.
"The protocols we've specified may seem pretty conventional, and indeed that's the point. Because they're so minimal, special case support isn't buried in the infrastructure, and developers must instead rely on what we call behaviors to extend the protocols for all interesting functionality," said Wilensky. In an open source environment, this approach allows for very flexible extension, and the researchers have already been able to implement a number of advanced capabilities, including such non-obvious ones as "lenses," which generate alternate views of selected areas, including OCR and magnification (Figure 2).
Wilensky and UC Berkeley computer science post-doc Thomas Phelps have developed a number of other behaviors including distributed annotation with copy editor marks, notes, highlighting, and hyperlinks, in which it is possible to add any of these annotations to someone else's web page in a new document, and then share this result with others, all without any special server support. "You can much more fluidly move through and interact with both your own information and other people's information. The boundaries are in some sense greyer, and collaborative work can emerge more spontaneously," said Wilensky.
The open-ended or "messy" nature of the Web with its dangling pointer, which many initially saw as a major potential problem, has opened opportunities for broader collaborations, which Wilensky wants to harness through this new model of multivalent documents. But these new freedoms bring new challenges. A central issue is how to make the new composited information in distributed annotations persistent, since it now inherently involves links to distributed information that the author does not control and that changes chaotically. The researchers are pursuing successful approaches to this robust hyperlink problem and the related subdocument reference problem, as well as issues of security and quality control or peer review.
Top | Contents | Next
PEER REVIEW THROUGH COLLABORATIVE FILTERING
The primary paradigm shift Wilensky believes digital technologies promise for scholarly communication is to move from a centralized, discrete publishing model, to a continuous, distributed, self-publishing model. Rather than submit papers to conventional journals and buy back the edited results at ever-increasing cost, scholars can self-publish on local resources, and then engage in a continuous, collaborative peer-review system.
"Academic communication is gated by technology in the sense that in the paper world paper is expensive so it pays to filter first. But in the electronic world there's no initial cost, so you can distribute first and filter later," said Wilensky. At first it might seem that peer review could be lost in such a model. But Wilensky and graduate student Tracy Riggs are finding that digital technologies offer novel opportunities for enhanced collaboration and peer review.
"Rather than being eliminated, scholarly review can take the form of distributed annotations and the use of reviewers or recommender systems," said Wilensky. Not only will academics easily be able to access primary resources, but they can create new composite documents containing annotations on resources, which they then rapidly share with others. Thus, the distributed digital library becomes a medium for continuous scholarly collaboration.
But how do you know what is good research amid the vast array of instantly published work? "You'd like to look at those articles that are given good reviews by good reviewers-those who have given useful and reliable reviews in the past as indicated by agreement with other reviews, and maybe looking backwards with the citation index. We've used a hubs and authorities type of algorithm to establish credentials," explained Wilensky.
The researchers are using this approach to build a collaborative filtering system, and in this way multivalent documents offer "something much finer grained than collaborative peer review or filtering that just makes an overall judgment," said Wilensky.
Beyond the GIS Viewer and multivalent documents, the researchers are investigating a wide array of other approaches, including such technologies as robust references; image retrieval by image content; document image analysis; distributed search; and natural language processing for information access. These technologies promise to extend the digital library of the future, enhancing the scholarly information life cycle. -PT
Top | Contents | Next
Thomas A. Phelps and Robert Wilensky. June 2000. Multivalent Documents. Communications of the ACM 43(6): 83.
Thomas A. Phelps and Robert Wilensky. July/August 2000. Robust Hyperlinks and Locations. D-Lib Magazine 6(7/8).