Coming Soon to Your Favorite Library:
Decision Support on Demand
Adapted for D-Lib Magazine
by
Hemant Bhargava and Bob Norris
U.S. Naval Postgraduate School
renorris@nps.navy.mil
This article is based upon published material from an ongoing, collaborative research project directed by Hemant Bhargava at U.S. Naval Postgraduate School, Ramayya Krishnan at Carnegie Mellon University and Rudolf Müller at Humbolt University.D-Lib Magazine, June 1996
ISSN 1082-9873
For the individual or organization wishing to
employ a scientific approach in solving decision problems, there is an abundance of relevant concepts,
methods, models, and software. Yet, decision technologies are little used in real-world decision making.
We argue that this disconnect is fundamentally rooted in the use of conventional market mechanisms to
distribute decision technologies and propose an alternative based on electronic markets.
After examining the problems confronting both consumers and providers of decision technologies, this
article explores how features of modern information networks, particularly those associated with the
Digital Library movement, can overcome these challenges. Finally, we introduce
DecisionNet, a new web-based marketplace for decision technologies.
Scenario
Problems Confronting Consumers
Problems Confronting Providers
Solutions Facilitated by Electronic Markets
DecisionNet
Digital Libraries and Decision Support
Project Status
Acknowledgements
Scenario
The following vignette portrays the role a remotely accessible decision technology can play in solving a challenging problem.
"Carolyn, we've set up a special Board of Director's meeting for you to demonstrate your decision making strategy to the whole crew. Again, please accept our congratulations for your work on the Billings Project. BTW, you're invited to dinner with the Boss, he's got some very good news for you."
With a laugh she flipped off the monitor and put her feet up, shaking her head as she remembered how it had all come together just a couple days before.
The Billings Project was known throughout the department as a can of worms. Corporate had acquired the small manufacturer at what seemed like a remarkable price, only to discover that their self-declared ability to meet delivery dates was a facade. Strapped with both exorbitant shipping and storage costs, due to hazardous material handling requirements, and confronted by a frustrated, widely-distributed customer base, Billings was approaching meltdown. Several heads at Corporate were on the chopping block and, as usual, it was flowing downhill.
As Carolyn watched her colleagues struggle to find a solution, she couldn't help but wonder whether some other company had faced a similar problem and created some type of decision technology to cope with it. While the tiger-team, led by Eric, were burning the midnight oil trying to model the distribution planning problem, she decided to do some research and logged into her favorite Digital Library (DL). Within minutes of her query, the Research Librarian provided a pointer to DecisionNet (D-Net), a Web-based service that opened the door to an electronic marketplace of decision technologies.
Following the prompts, Carolyn answered a few basic questions and requested assistance in choosing relevant technologies. DecisionNet analyzed her requirements and, from its yellow pages, identified four candidate decision technology providers. Carolyn was surprised to learn that D-Net also provided format translation and actually facilitated the interconnection between her web browser and each provider's system. Even more exciting was the discovery that two of the providers offered free trial runs.
Grabbing the working dataset from one of her harried co-workers, Carolyn took a deep breath and logged onto the first system. She found the user-interface intuitive and was soon immersed in the process. Though she liked the system, she was concerned that this optimization model didn't seem to have an easy way to factor in the costs for each state's hazardous material transportation taxes. She left an e-mail describing her problem and moved on to the second provider.
Again, the user-interface was designed for a novice and even offered a JAVA-scripted tutorial. Carolyn was thrilled to find a customizable, geo-based cost function that allowed her to enter different tax rates for each state. She crossed her fingers and ran the working data-set. In seconds, a solution was displayed that beat Corporate's cost threshold by 22%. Knowing she had a winner, Carolyn followed the links to the provider's sales office, downloaded a user-rate table and walked into her supervisor's office.
The rest, as they say, is history.
Potential Consumers face the immediate
obstacle of defining their decision needs and choosing an appropriate solution technology. A host of
challenges complicate the issue. Within the context of the scenario, we will examine those relevant to the
underlying market mechanism, by comparing Carolyn's experience to the plight of her co-workers.
1. Awareness: Users may not be aware of relevant technologies.
To successfully attack the problem, Eric's tiger-team had to know that they were faced with a distribution
planning problem that could be formulated as an optimization model. This level of expertise might well be
absent in many organizations. The implementation of decision technologies in real-world problem solving
is hindered by scant awareness among potential users.
With a market integrated into the realm of Digital Libraries, Carolyn was able to follow the path of a
traditional information seeker, that is, she contacted a Research Librarian and asked for help. After being
referred to DecisionNet, Carolyn's first interaction was with a built-in expert system. Responding to
system prompts, Carolyn did her best to define the problem. The expert system responded by classifying
her information needs and, after asking Carolyn to confirm its analysis, the system made specific
recommendations.
2. Accessibility: Users may not have access to beneficial technologies.
Armed with the shipping data Eric's group had assembled, Carolyn was guided to key decision technology
components, that is, models and modeling environments.
Without D-Net, the problem would have been far more difficult. Eric's group would have to identify
publishers and vendors of each relevant technology or be forced to adapt software they had previously
acquired or inherited from other projects. Search costs currently associated with accessing decision
technologies are especially high and crisis programming, particularly for one-shot applications, can be
costly, disruptive and/or unproductive.
3. Compatibility: Most technologies require specific hardware and software configurations.
Under the status quo, decision technology acquisitions must be compatible with the Consumer's IT
architecture. Thus, a modeling environment only available on Windows NT, for example, would be
incompatible with Eric's Macintosh platform. This problem is also found in the context of platform
configurations. A model may be programmed using features available in a certain release of the modeling
environment software which in turn may be limited to a certain release of the operating system.
For Carolyn, once she was connected to DecisionNet, compatibility was not a concern.
4. Applicability: Due to both the complexity - expertise, effort, and cost - of developing decision
technologies and the limited market-base, there is little motivation for providers to create easily adaptable
models.
Carolyn was able to choose from four candidate providers because the expanded market horizon offered by
the Digital Library system encouraged competition and innovation.
5. Interoperability: Many decision problems require a combination of technologies to provide a
satisfactory solution.
For the Board of Director's meeting, Carolyn might have to extract problem-specific data from a database
server, cast it in the format of the modeling language, submit it into the modeling environment for
execution, and then present the results graphically. Since each of the technologies has its own input and
output formats, data interchange to enable interoperability is an important requirement.
Under the status quo, interoperability requires either substantial human intervention or integration of a
variety of tools on a single platform (see compatibility problems).
Among the challenges faced by Providers,
those that relate to the market are symmetric to the problems encountered by Consumers.
1. Advertisement (awareness & accessibility):
As new decision technologies are developed, providers need to attract users. Currently, it is difficult for
specialized software providers to cost-effectively reach consumers.
2. Heterogeneity (compatibility): Even in niche markets, there is
heterogeneity of computational platforms. For providers looking for market share, this means the expense
of supporting the technology on a variety of platforms.
3. Version Management (compatibility): Often a working version of a
product is rendered useless due to a change in, say, operating system software. Even in the absence of
shifts in the user platform, a variant of this problem is encountered as there is a need to upgrade and
maintain software over time.
4. Customization (applicability and
interoperability): The cost of producing and customizing decision technology
software
using traditional software distribution strategy is high due to the small and specialized nature of the
market. This is exacerbated by the need to offer coordinated or integrated interoperable solutions.
Any organization would have to overcome all the consumer problems before they could apply the decision
technology-based solution Carolyn located using DecisionNet. For the occasional user, this can be a costly
process. For the novice, it can be downright intimidating. Without the expanded market of DL users, it is
doubtful that those four providers could have afforded to engineer intuitive user interfaces and offer
adaptable customization.
Treating decision technologies as
information products or services that can be accessed over an information network is the fundamental idea
behind DecisionNet. We propose the development of an electronic market in which consumers and
providers transact access to decision technologies. Summarized below are the enabling features of the Web
and the Internet that address the consumer and provider problems discussed above.
1. Global Hypermedia Information System: Awareness; Accessibility; Applicability;
Advertisement; Customization
The WWW offers an excellent opportunity to address, via creation of globally accessible "yellow pages,"
the awareness and advertisement problems-it has
enormous reach, is easy to use, and electronic search can be convenient and powerful. A yellow pages
market would include a listing of, and information about, available decision technologies in particular
categories.
Benefiting from being DL compliant, such a market would also have a classification scheme for
technologies and features to locate, via indexed search and retrieval, suitable products or providers. The
potential market of DL users would have an indirect effect on the applicability
and customization problems-by lowering advertising and distribution costs and
encouraging the development of specialized or niche products.
Further, due to its protocols for data transport between heterogeneous networks, the Internet can also
address the accessibility problem; technologies on the yellow pages can be linked
to the actual software that can be downloaded or distributed over the Internet.
2. Common Gateway Interface: Compatibility; Awareness; Heterogeneity, Version
Management
The Common Gateway Interface (CGI) to executable programs can be used to allow users to remotely
execute decision technologies. Like Carolyn, users can input computational parameters via HTML forms
right in their Browser. Remote execution completely solves the compatibility
problem since the user never needs to own or install a copy of the software. From the provider's
perspective, it also solves the heterogeneity and version
management problems.
The remote execution feature is also required to support indexed search of yellow pages, which helps users
locate and become aware of appropriate decision technologies.
Another approach to the same set of problems would involve using Sun Microsystems' Java language.
Java executable programs (applets) can be downloaded to users' machines to be executed on the client's
processor without requiring compilation or installation.
3. Distributed Nature: Applicability; Interoperability
The distributed nature of the Web means not only that several consumers can use products on remote
machines, but also that multiple providers can enter the market supplying similar technologies. Further,
consumers may be able to interconnect multiple technologies - possibly from different providers - in ways
that would not otherwise be possible.
In the scenario, Carolyn interacted with a single provider. This exchange could also have been facilitated
through an interoperative market where, for example, the working data set would be transposed by a
formatting service into GAMS syntax, combined with a transportation model supplied by an
independent provider, and executed in a modeling environment owned by yet another. This support for interoperability invites providers to participate in the market even if they only
provide a piece of the solution and enables consumers to flexibly design their own solutions.
DecisionNet aims to establish a
marketplace which implements features of both information and execution markets. Listings of registered
technologies are organized into yellow pages which can be searched. The entries are hyperlinks to the
technologies and encode the appropriate access semantics (e.g., ftp for download, telnet for execution).
These entries are created as part of a technology registration process which also automates much of the
overhead of creating a web-accessible technology. Key features of DecisionNet include:
1. DecisionNet features all types of decision technologies from data sets to modeling environments,
placing the niche data-set provider on equal footing with those that own high-end, expensive
computational platforms and modeling environments.
2. Providers are given access to an intelligent registration agent which leads them through a series of
steps resulting in both a listing for DecisionNet yellow pages and in the automated creation of a web-based
user interface to the technology.
3. Consumers in DecisionNet can use the services of an intelligent agent to cobble together a solution
using available technologies.
4. Providers are not required to supply their technologies on a platform owned by DecisionNet. After
registering the appropriate protocol to invoke their technology (e.g., anonymous telnet or the POST
method of the http protocol), providers maintain their own servers, leveraging the distributed nature of the
web and permitting scalability.
DecisionNet technologies are developed and maintained locally by their providers.
The DecisionNet
server contains the meta-information necessary to guide consumers in search, selection, and execution of
these technologies. The system is organized on a "pay-per-use" (though, at this time, it is
non-commercial) concept that treats decision technologies as "services" that are
used rather than as products that are owned by users.
Independent technologies are those for which you must craft the WWW- interface and execution yourself; these technologies can therefore also run independently of DecisionNet. You must also program overhead functions such as user registration and accounting, authentication, and billing. To supply such technologies in the DecisionNet environment, you register them by providing basic meta-information such as purpose, location, and classification information. To provide such a technology, you must have a WWW server, possibly an FTP server, and run some CGI programs or equivalent.
Exclusive technologies are those for which DecisionNet agents provide the WWW-interface, execution control, and overhead functions; these technologies must therefore run exclusively under DecisionNet. To supply such technologies in the DecisionNet environment, you register them by providing basic meta-information as well as criteria, for example format and priority, on the inputs and outputs that DecisionNet agents must address when setting up the WWW interface. The decision technology must be available to DecisionNet agents via a port on a telnet or WWW server.
Integrating decision support technologies
into the Digital Library movement is an exciting and timely concept that can bring powerful
computational technology to the user's fingertips.
The concepts of decision support have been around for over 25 years - longer than the time it took
database systems to become a permanent and indispensable tool in today's organizations. Despite
demonstrated utility and need, the mechanisms haven't existed to mate the occasional or non-technical
user with a suitable provider. Without a market, the field has languished in relative obscurity and limited
funding. Yet, decision support technologies are uniquely capable of providing solutions to difficult
challenges. With certainty based upon economic principles, we assert that providers will respond to
electronic market based demand for functionality and adaptability with enthusiastic innovation and
healthy competition.
Digital Libraries are responding to user demands for access to information in all its forms. Given ever-
increasing access to resources, both technology consumers and Librarians face the challenge of extracting
information from a morass of data. Thankfully, connectivity and interoperable systems can offer far more
to the user than mere access to data. By opening the door to remote computational systems, the DL
community has the opportunity to greatly enhance library services and jump-start an entire field of
scientific work.
A prototype of DecisionNet is currently
on-line. Though some aspects of the functionality portrayed in the scenario, for example the front-end
expert system, are still in development, we offer a demonstration of the technology for your inspection.
Visit DecisionNet |
---|
DecisionNet is a collaborative research project involving faculty and students at the Naval Postgraduate School (NPS, Monterey), Carnegie Mellon University (CMU, Pittsburgh) and Humboldt University (HU, Berlin).
The principal investigators directing the project are Hemant Bhargava and Ramayya Krishnan. We have strong links to the MMM system at HU via Dr. Rudolf Müller and others working on MMM.
This summer (1996), we will expand the yellow pages and enhance the user-interface. If you are a Decision Support Provider, a potential research sponsor, or wish to comment on our work, address your e-mail to:
While adapting the work of the principal investigators into an article suitable for D-Lib Magazine, significant portions of these two works were used without direct attribution:
H.K. Bhargava, R. Krishnan, and R. Müller. Their joint paper: Decision support on Demand: Emerging Electronic Markets for Decision Technologies, February, 1996.H.K. Bhargava, R. Krishnan, and R. Müller. On parameterized models for agents in electronic markets for decision technologies. In Proceedings of the Fifth workshop on Information Technologies and Systems, Amsterdam, Holland, December 1995, pages 218-227, 1995.
The DecisionNet Team acknowledges the following funding:
National Science Foundation (grant IRI-9312143), the U.S. Army Artificial Intelligence Center (grant MIPR6GNGS00087), U.S. Army Research Office (grant DAAH04-94-G-0239), the Institute for Joint Warfare Analysis at the Naval Postgraduate School, and the National Research Center SFB 373 of the Deutsche Forschungsgemeinschaft.
Created by Bob Norris, U.S. Naval Postgraduate School, 1 June 1996.
hdl://cnri.dlib/june96-norris