See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225898083

Model Management and Solvers for Decision Support

Chapter · January 2008

DOI: 10.1007/978-3-540-48713-5_12

CITATIONS READS

8 4,311

5 authors, including:

Ting-Peng Liang Ching-Chang Lee

National Sun Yat-sen University National Kaohsiung University of Science and Technology (NKUST)
225 PUBLICATIONS   9,667 CITATIONS    5 PUBLICATIONS   327 CITATIONS   

SEE PROFILE SEE PROFILE

Efraim Turban Frada Burstein

University of Hawaiʻi at Mānoa Monash University (Australia)
199 PUBLICATIONS   13,337 CITATIONS    257 PUBLICATIONS   3,056 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Self-sustaining Regions Research and Innovation Initiative View project

Supplier Development View project

All content following this page was uploaded by Ting-Peng Liang on 21 March 2016.

The user has requested enhancement of the downloaded file.

CHAPTER 12
Model Management and Solvers
for Decision Support

Ting-Peng Liang1, Ching-Chang Lee2 and Efraim Turban3

1
National Sun Yat-sen University, Kaohsiung, Taiwan
2
National Kaohsiung First University of Science and Technology, Kaohsiung, Taiwan
3
Pacific Institute for Information System Management, University of Hawaii, Manoa, HI,
USA

In this chapter, we examine a major function in class of decision support systems (DSS):
model management in model-oriented DSSs. In particular, the chapter focuses on the con-
cept of model management systems and explores key issues in model management: re-
presentation, formulation, composition, integration, selection, and implementation. It also
reviews various kinds of solvers commonly available for decision support and the growing
importance of Web Services technologies for delivering model/solver management cap-
abilities.

Keywords: Decision support system; Decision model; Model management; Solver; Web
Services

1 Introduction

An important feature of decision support systems is that many, perhaps most,

include at least one decision model. A model is an abstraction of a specific problem
or a class of problems in the real world. Models are useful tools for managers in
solving problems or making decisions. Models can help decision-makers under-
stand problems and help them make decisions. The purpose of a model-based, or
model-driven, decision support system (DSS) is to provide decision-makers with
useful decision models for analyzing and solving complex problems.
A DSS whose key value is the powerful model included in its model base is
called a model-oriented DSS. For instance, a transportation company relies heavily
on the routing of its vehicles in order to deliver customer packages efficiently. The
vehicle routing DSS is a model-oriented DSS. A model-oriented DSS is different
from a data-oriented DSS in that the decision models adopted in the system is more
complicated and plays a more critical role toward the final decision.
A “model” can be a graphic or a set of mathematical formulae used to explain
the relationships among relevant variables of a particular problem. For example,
232 Ting-Peng Liang et al.

a linear programming model for finding an optimal solution to a product-mix

problem involving two products and one ingredient is represented as follows:

Maximize:

Profit = p1X1 + p2X2

Subject to:

a1X1 + a2X2 <= Qa

X1 <= D1

X2 <= D2

Where:
Qa is the amount of Ingredient A allocated to produce Products X1 and X2;
Unit Profits for X1 and X2 are p1 and p2, respectively;
Market Demand for X1 and X2 is D1 and D2, respectively;
Unit Amount of A required for producing X1 and X2 is a1 and a2, respectively.

When we have specific data values for a model’s parameters (e. g., for Qa, p1, p2, D1,
D2, a1, a2 in the example above), then the model is instantiated to form a problem
statement. That is, the model coupled with a particular data set amounts to a specific
problem that we can try to solve (e. g., the problem of finding particular values of X1
and X2 that maximize profit in the context defined by the data values for the model’s
parameters). Software that can solve the stated problem is called a solver; it is a data
analysis procedure that derives a solution for any correctly modeled instance of
a class of problems. For example, a linear-programming (LP) solver automates an
algorithm for finding a solution to any LP problem—comprised of an LP model and
a data set.
Thus, the study of model management aims to understand the possibilities for
computer-based treatment of three key elements:
• Models;
• Associated data sets;
• Related solvers.
Within a system that aids decision makers by solving problems they encounter in
the course of reaching decisions (Basu and Blanning, 1994; Chang et al., 1993). In
DSS parlance, it is common to see “models” used to refer to computer programs
(i. e., solvers) that use representations of relationships among variables to help find
solutions to complex problems. That is, the term is often understood to include
solvers that can process the models. For example, a DSS for production scheduling
needs to have a scheduling model and solver that can operate on that model for any
 Model Management and Solvers for Decision Support 233

selected data set in order to determine the optimal production sequence for the
selected context.
Many DSS researchers contend that the study of models and their management
is a key ingredient for developing successful DSSs (Alter, 1980; Blanning et al.,
1992; Blanning, 1993; Bonczek et al., 1980, 1981a; Elam, 1980; Haseman et al.,
1976; Keen and Scott Morton, 1978; Liang, 1988; Sprague and Carlson, 1982; Stohr
and Tanniru, 1980). An empirical study reports that decision makers with model
management system (MMS) support outperformed those without the support
(Chung et al., 2000).
The purpose of this chapter is to describe the development of model management
systems and related DSS research issues. The remainder of the chapter is organized
as follows. First, basic concepts of models and the modeling process are described
in Section 2. Model management functions and system architecture are provided in
Section 3. Key issues in model management are reviewed in Section 4. Solvers that
calculate outputs for models are examined in Section 5. Web-based model man-
agement and new research directions for the future are discussed in Section 6.

2 Models and DSS Modeling

A model-based, solver-oriented DSS is designed to leverage models, data, and user

interfaces to help decision makers. Most models used in such DSSs are mathe-
matical models that have a target output, a set of inputs, and operations for con-
verting inputs to outputs (Liang, 1988). A mathematical model describes the
modeled problem by means of variables that are abstract representations of those
elements of the problem that need to be considered in order to evaluate the
consequences of implementing a decision. Figure 1 illustrates the structure of use of
a model for decision support:

Figure 1. Structure of a mathematical model for decision support

234 Ting-Peng Liang et al.

• Decision variables (input) x, are those controlled by the user;

• External variables (input) y, are parameters not controlled by the user
(e. g., determined by the environment and context of a problem);
• Outcome variables (output) z, used for measuring the consequences of
implementation of inputs;
The relationship between inputs x and y, and outcomes z are presented in the form:
Z = F(x,y).
The purpose of a DSS is to support a user in finding values for his/her decision
variables x that will result in an optimal, or at least good enough, solution to the
problem (Makowski, 2004). Mathematical models can be categorized into three
classes: Prescriptive Models, Predictive Models, and Descriptive Models. In any
case, they may provide substantial benefits to decision makers and must be
managed properly in DSSs.

2.1 Benefits of DSS Models

Using models in DSSs has many advantages. The major ones are (per Turban et al.,
2007, p. 52):
• Models allow easy manipulation of decision variables (changing data
values of decision variables or environment variables) to investigate
potential scenarios, which is much easier than manipulating the real
system. Experimentation with and on models does not interfere with the
daily operation of the decision maker’s organization.
• Models enable the compression of time. Years of operations can be si-
mulated in minutes or seconds of computer time.
• The cost of modeling analysis is much less than the cost of a similar
experiment conducted on a real system.
• The cost of making mistakes during a trial-and-error experiment is much
less for using models than in the real world.
• Models can help estimate and reduce risks. The business environment
involves considerable uncertainty. With modeling, a manager can esti-
mate the risks resulting from specific actions and develop contingency
plans.
• Mathematical models enable the analysis of very large, sometimes infi-
nite, numbers of possible solutions. Even in simple decision problems,
managers often face a large number of alternatives. Decision models can
substantially simplify the analytical process.
• Models enhance and reinforce learning and training.
• Models and solution methods are readily available for solving some
typical problems. Their proper use can enhance decision performance.
• There are many solvers, including Java applets (and other Web pro-
grams), that can be used to solve models.
 Model Management and Solvers for Decision Support 235

2.2 The Modeling Process

Modeling is a process of creating, representing, evaluating, and documenting a de-

cision model. A typical modeling process includes the following steps:
• Define the scope and specification of the problem. The model builder
who develops a model needs to determine the scope of the model
by identifying key parameters and variables related to the decision
problem.
• Identify the nature of the defined parameters and variables. It is essential
that the domain and value range of variables must be determined. Variables
may be deterministic (i. e., having a specific value) or probabilistic (values
fall in a probability distribution).
• Construct the mathematical relationship of variables. The relationship
may be known in advance (e. g., calculating the total cost of a product) or
need to be determined through data analysis (e. g., regression analysis).
• Implement the mathematical model in a computer-based environment and
verify whether the program follows the specification.
• Validate the computer models with empirical data or logical reasoning.
• Repeat the above procedures until the model accurately reflects the
reality.
Krishnan and Chari (2000) argue that model development is a complex, iterative
process during which several modeling tasks need to be accomplished. With the
exception of model administration, the tasks may be broadly classified as
contributing to either the pre-solution, or solution, or post-solution phases. They
outline a few major tasks in a modeling lifecycle, as illustrated in Table 1.
In order to facilitate the modeling process, a few frameworks have been
proposed. These modeling frameworks specify how concepts and relationships
involved in a model can be represented and manipulated. For instance, Geoffrion
(1987, 1989, 1992a, 1992b) has advanced the structured modeling approach. Others
include the models as data view (Bonczek et al., 1976; Lenard, 1986), relational
framework (Blanning, 1982, 1985), the entity-relationship framework (Blanning,
1986; Elam, 1980), object-oriented approaches adapted from data management
(Bonczek et al., 1983; Huh and Chung, 1995; Lenard, 1993; Pillutla and Nag,
1996), a problem reduction approach (Bonczek et al., 1979), first-order predicate
logic (Bonczek et al., 1981a, 1981b; Dutta and Basu, 1984), and other artificial
intelligence approaches (Dolk and Konsynski, 1984; Elam and Konsynski, 1987).
Graph-based approaches, which borrow concepts from the graph theory for
representing mathematical models, are also useful (Jones, 1990, 1991; Liang, 1988;
Basu and Blanning, 1994, 1998).
236 Ting-Peng Liang et al.

Table 1. Tasks in the modeling life cycle. Source: Krishnan and Chari, 2000

Task Goal Mechanism

Problem
Clear, precise problem statement Argumentation process
identification

• Formulation
Statement of the model(s) required • Integration
Model creation to mathematically describe the • Model selection and
problem modification (if necessary)
• Composition

• Ad hoc program development

• Use of high-level specialized
Model Computer executable statement of
languages
implementation the model
• Use of specialized model
generator programs

Symbolic analysis of attributes

Model validation Feedback from validator such as dimensions and units
syntax rules

• Solver binding and execution

Model solution Feedback from solver • Solver sequencing and control
script execution

• Model comprehension
Model • Structural analysis
• Model debugging
interpretation • Sensitivity analysis
• Model results analysis

Model Revise problem statement and/or Symbolic propagation of

maintenance model to reflect changes/insight structural changes
Maintain correct and consistent
Model • Versioning
versions of models. Ensure
versions/security • Access control methods
authority to access.

3 Model Management

3.1 Model Management Concepts and Benefits

As models play very important roles in human decision-making processes, they are
considered as a valuable organizational resource that should be managed properly.
Therefore, development of model management systems has been one of the most
 Model Management and Solvers for Decision Support 237

important research areas for decision support systems (Liang and Jones, 1988).
A MMS is a software system that provides tools and environments to develop,
store, and manipulate models, data, and solution methodologies associated with
complex decision problems (Gnanendran and Sundarraj, 2006).
There are a few major reasons for model management. First, model development
is a complicated and expensive process. A good MMS can increase the reusability
of existing models. Second, sometimes a model can be built from existing com-
ponents. An MMS can also enhance the productivity in a modeling process by
providing relevant components (also called building blocks or modules in some
literature). Third, different units in an organization may need similar models and
build them separately. An MMS can avoid redundant effort and increase model
consistency in the same organization through model sharing. Fourth, an MMS can
provide a better integration between models and their data. Finally, an MMS can
provide model developers and users with better documents and maintenance
support.

3.2 The Architecture of a Model Management System

In general, a DSS’s model management system includes the following elements, as

illustrated in Figure 2 (Liang, 1988; Turban et al., 2007):

Figure 2. Framework of model management systems

238 Ting-Peng Liang et al.

• Model base: A model base is a collection of computer-based decision

models. Its function is similar to a database, except that the stored objects
are models. The models in the model base can be divided into different
categories, such as strategic, tactical, operational, and analytical.
• Model base management system (MBMS): A MBMS is software that
handles the access of model base and the linkages with other com-
ponents. It often includes a model development environment (MDE) and
a model execution environment (MEE) for handling the models in the
model base.
• Model directory: The role of the model directory is similar to that of
a database directory. It is a catalog of all the models and other software in
the model base. It contains model definitions, and its main function is to
answer questions about the availability and capability of the models.
• Model development environment (MDE): A model development envir-
onment supports model builders to construct useful models. It needs to
include a model definition language (MDL) that allows models to be
represented properly and save within the model base for execution. It also
provides a platform on which models can be created, saved, integrated,
selected, and maintained if necessary.
• Model execution environment (MEE): A model defined in a definition
language can be executed in the MEE. It includes a model manipulation
language (MML) that executes existing models to obtain an optimal
solution. It also has an interface for the user to manipulate the selected
model and links to the solver and the data management modules.
• Solvers: A solver is a software tool that helps users to manipulate models in
order to find a solution (e. g., optimal) to a stated problem, by following
some definite procedure. For example, linear programming is the best-
known technique to solve resource allocation problems. An LP solver,
such as LINDO which implements the simplex algorithm, can help the
decision-maker to find the best way to allocate scarce resources (Lindo
System Inc., http://lindo.com/company/index.html). A “best” or “optimal”
solution may mean maximizing profits, minimizing costs, or achieving the
best possible quality. More information about solvers is provided in
Section 5.
These elements are related to a DSS’s relevant data and knowledge management
abilities and are accessed via the DSS’s user interface.

3.3 Functions of a Model Management System

The model management activities are analogous to those of the database

management system (DBMS) and they include the following activities and tasks of
managing models:
 Model Management and Solvers for Decision Support 239

1. Representing a model: to specify a model accurately.

2. Constructing new models: to provide an environment to help deci-
sion-makers develop new models conveniently. This is the essential
function in MDE.
3. Integrating models: to create a larger model by combining existing
models or building blocks.
4. Storing models: to save existing models in the model base.
5. Accessing models: to facilitate the retrieval and use of decision models in
the model base.
6. Selecting models: to select existing models from directories and libraries.
7. Executing models: to execute an existing model and present outputs to
decision-makers.
8. Maintaining models: to update and modify models in the model base.
9. Allowing for ad hoc manipulations: to enable users to manipulate models
so that they can conduct experiments and sensitivity analyses ranging
from what-if to goal-seeking analyses.
10. Indexing models: to provide catalogs and displaying the directory of
models for use by several individuals in the organization.
11. Tracking model data and application use.
12. Linking models and data for execution.

4 Key Issues in Model Management

Researchers have investigated several key issues in model management, including

model representation, model formulation, model composition, model integration,
model selection, and model implementation. These are elaborated below.

4.1 Model Representation

A model is comprised of a set of inputs, outputs, and operations that convert from
inputs to output. Therefore, the basic representation of a model needs to include
these three dimensions. In addition, certain information affiliated with model man-
agement such as integrity constraints and validity information may also be useful.
There are many ways in which a model may be represented for storage and
utilization. Early approaches treated models as data relations or objects, and defined
attributes associated with these objects (Bonczek et al., 1982, 1983). Later, the
structured modeling framework developed by Geoffrion proposed five basic ele-
ments for model representation: primitive entity, compound entity, attribute, func-
tion, and test. These elements are grouped by similarity into any number of classes
called genera, and organized hierarchically as a rooted tree of modules so as to
reflect the model’s high-level structure. The dependencies among these elements
240 Ting-Peng Liang et al.

can be represented as arcs and illustrated graphically in a directed, acyclic graph,

called the model structure, which has three levels of detail, called elemental,
generic, and modular structures (Geoffrion, 1987).
Models may be represented at different levels: model instance, generic model,
and model paradigm. A model instance is a particular model used for solving
a particular decision problem in an organization. All parameters are set for the
particular instance. For example, Exso Corporation may run a truck routing model
to arrange its shipping for the coming Monday. A generic model is a class of model
instances whose parameters are known but not yet instantiated. The truck routing
model tailored for Exso is a generic model. The model may belong to the linear
programming model paradigm that can be solved by the Simplex method. This
classification helps our understanding of models at different levels of abstraction.
Other model representation approaches that have been proposed include models
as data (Bonczek et al., 1976; Lenard, 1986), models as frames (Dolk and
Konsynski, 1984; Elam and Konsynski, 1987), the object-oriented approach (Huh
and Chung, 1995; Pillutla and Nag, 1996), the graph-based approach (Jones, 1990,
1991; Basu and Blanning, 1994) and Unified Modeling Language representation
(Hong and Mannino, 1995; Dolk, 2000).
Because models are built, managed, and applied by persons with varying
cognitive skills, Greenberg and Murphy (1995) argue that multiple representa-
tions of a model are important to accommodate these variations. They suggest
a multi-view architecture that allows different users to readily gain insights into
problem behavior from their own respective perspectives (e. g., that of model
builder vs. problem owners). Their formal framework for this architecture includes
model views involving algebraic, block schematic, graphic, and textual repre-
sentations, with merits of alternative representations being determined by the extent
to which each aids in comprehension and insight.

4.2 Model Formulation

Model formulation research focuses on converting a precise, problem description

into a mathematical model. It includes two lines of research: one is to provide an
environment to facilitate manual model formulation and the other is for automated
model formulation.
Murphy and Stohr (1986) implemented a system for writing linear programs
(LP) with artificial intelligence (AI) techniques. In this research, AI is used to
simplify the problem formulation process. Ma et al. (1989) introduce a new repre-
sentation that allows modelers to depict their problems in a graphical rather than
mathematical form. This representation is described in detail together with several
interface design principles that will aid modelers—including hierarchical decom-
position, multiple model representations, alternative formulation approaches, the
use of model templates, and database and model management facilities. In later
research, Ma et al. (1996) use textual and graphical means for problem represen-
tation and implement a LPFORM model formulation system.
 Model Management and Solvers for Decision Support 241

Binbasioglu and Jarke (1986) were the first to propose an alternative domain-
specific approach to knowledge-based model formulation problems. The domain-
specific approach has since been elaborated in the work of Krishnan (1990; 1991),
Bhargava and Krishnan (1992), and Jones and Krishnan (1992). Raghunathan et al.
(1993; 1995) also elaborated the domain-specific approaches on model formulation
problems.
Jones (1990, 1991) proposed a graph-based model system (GBMS) to facilitate
the construction of models using attributed graphs. A set of graph-grammars was
developed to construct different types of models, and a prototype GBMS system
implemented. Chari and Sen (1997) also proposed a Graph-Based Modeling System
for Structured Modeling (GBMS/SM) using acyclic graphs to represent models in
various domains.
Liang and Konsynski (1993) and Liang (1993) proposed a different approach
based on analogical reasoning. They used a tree-like hierarchy problem repre-
sentation and used a case-based reasoning approach to identify analogies between
a problem description and a previously formulated case. Tseng (1997) used
a blackboard control approach to support diverse reasoning behaviors during model
formulation.

4.3 Model Composition

Model composition leverages previously developed models, but does so by linking

together independent models such that the output of one model becomes an input to
another. Model composition is often used in conjunction with model selection when
no one model meets the requirements of a problem. A distinguishing feature of
model composition is that none of the individual models that are linked are
modified after they have been selected. An example of model composition is the
linking together of a demand forecasting model and a production scheduling model,
such that the forecasted demand is used as a parameter in the production scheduling
model (Krishnan and Chari, 2000). Research on model composition can be
classified as relational, graphical, script-based, and knowledge-based approaches
(Chari, 2002).
In the relational approach, models are treated as virtual relations. A linkage
between two models is specified using a model join operator (Blanning, 1982,
1985; Bonczek et al. 1983).
In graphical approaches, models are viewed as nodes or edges of a graph. Liang
(1988) proposes the use of automated model composition using AND/OR graph
search. Basu and Blanning (1994) propose using metagraphs to facilitate model
composition. Muhanna and Pick propose a systematic approach that adopts
meta-modeling concepts for capturing the semantics of a modeling process to
support model composition (Muhanna, 1993; Muhanna and Pick, 1994).
In script-based approaches, model composition is performed via predefined
scripts. For instance, Jeusfeld and Bui (1997) propose a uniform naming and a data
242 Ting-Peng Liang et al.

representation scheme to facilitate model composition over the web. Several

modeling languages such as GAMS use scripts to represent models and composite
model.
Knowledge-based approaches focus on using rules and reasoning mechanisms
for combining models. For instance, Bonczek et al. (1981a, 1981b) and Dutta and
Basu (1984) use predicate logic to represent models, and to effect model
composition. Each model is a predicate with some of the arguments being treated as
inputs, and the others considered as outputs. Stored reasoning knowledge in the
form of first-order predicate expressions (i. e., rules) is drawn on by the DSS’s
problem processing system as it uses inference to compose a model (from extant
models) for solving a user’s request. The request does not need to identify or specify
how the models are to be combined with each other or with the appropriate data sets
for the user’s problem context. Liang (1988) developed a rule-based mechanism to
support the composition of several models from their linkages of inputs and outputs.
Chari (2002) proposes filter spaces to facilitate automation of model composition
and execution processes, and to integrate partial solutions from models and
databases.

4.4 Model Integration

Like model composition, model integration also leverages previously developed

models. However, in model integration, the models being integrated are modified.
Model integration is probably the most complex mechanism used to accomplish
model creation (Krishnan and Chari, 2000). Two broad sets of topics have been
identified as germane to model integration research. They are schema integration
and process (solver) integration (Dolk and Kotteman, 1993). Schema integration is
the task of merging the internal structure of two models to create a new model.
Process integration is similar to solver integration in the context of model
composition. A key distinction is that in model composition, the solution of any
given model is an independent process. In model integration, the solution process of
two or more models may have to be interwoven.
There are a host of issues that are relevant to model integration. For instance,
Bradley and Clemence (1987) propose a type calculus for modeling languages with
the specific objective of identifying conflicts. Krishnan et al. (1993) describe the
use of types and type inference in a language to facilitate conflict resolution as well
as schema integration. Bhargava et al. (1991) introduce a concept to detect conflicts
in names, type, and dimension.
Geoffrion’s (1987) structured modeling provides a formally specified notational
framework for modeling that was developed to address a variety of model devel-
opment problems. Geoffrion (1989; 1991; 1992) also describes a detailed manual
procedure for integrating models specified in the SML language.
Liang (1990) investigates the reasoning process in model integration. With an
eye on another issue, Dolk and Kotteman (1992, 1993) propose solver integration
 Model Management and Solvers for Decision Support 243

methods. They use the communicating sequential processes theory to enable

dynamic variable correspondence and synchronization in solver integration.

4.5 Model Selection

Model selection leverages the existence of previously developed models to create

a model for a new problem. It aims to select appropriate models for problem solving
from the model base. An advantage of this approach is the ability to reuse models
(Krishnan and Chari, 2000). Typical concerns in model selection are tradeoffs
between model performance, generalizability, and complexity (Foster, 2000;
Myung, 2000). Banerjee and Basu (1993) propose a methodology to systematically
direct users to obtain needed information and make tradeoffs for selecting model
type under given economic, technological, and other constraints. Research on
model selection from the model management perspective is limited and a fertile
area for future research.

4.6 Model Implementation

Model implementation is the task of creating a model representation to which

a solver can be applied (Krishnan and Chari, 2000). It is done either in a software
tool such as a spreadsheet or in an executable modeling language tailored for certain
types of models, such as LINDO for linear programs. Four guidelines for a model
implementation environment are proposed in Krishnan and Chari (2000):
model-data independence, model-solver independence, model-paradigm indepen-
dence, and meta-level representation and reasoning.
The most popular model implementation tools are spreadsheets. These tools do
not support model-data independence because each spreadsheet model is a model
instance. This is a shortcoming that limits the size of the model that can be
understood, verified, and maintained. However, spreadsheets do preserve mo-
del-solver independence. Multiple solvers can be applied to a model. Because only
model instances are specified in spreadsheets, model-paradigm independence is
also supported to a certain degree. For instance, spreadsheets allow optimization
models and regression models to co-exist and communicate.
Algebraic modeling languages are another important contribution of model
management research to improve modeling productivity. They allow models to be
specified and reused more effectively. Before the advent of these languages, ad hoc
programs and matrix generator programs had to be developed to implement
mathematical programming models.
Structured modeling language also provides a basis for model implementation.
For instance, Geoffrion (1991) developed a prototype system called FW/SM to
implement a structured modeling language environment. Gagliardi and Spera
(1997) also propose an object-oriented modeling language called BLOOM based on
244 Ting-Peng Liang et al.

a structured modeling framework. GAMS (http://www.gams.com/) and AMPL

(http://www.ilog.com/products/ampl/) are examples of commercial modeling
languages that support model implementation.

5 Solvers

Beyond model development, it is essential to solve a model efficiently. A solver is

basically a preprogrammed model-solving algorithm comprised of executable
routines that generate model solutions. Without proper solvers, the solution of
a model may not be found. Because DSSs have been used in many problem
categories and in many business and service areas, there exist several model
categories, with many sub-categories, for which solvers have been developed.
Hence, there are hundreds of solvers, some of which are commercially available;
others are proprietary in corporations, governments, or in universities.

5.1 Commercial Solvers

Commercial solvers can be divided into two major categories:

• Generic OR/MS solvers, which are briefly described in this section.
• Specialized solvers, which deal with specific functional fields (e. g.,
finance, marketing, production), specific industries (airlines, banking,
hospitals), or specific problems within industries.
A good source of information on commercial solvers is the OR/MS Today
periodical published by the Institute for Operations Research and Management
Science.

5.1.1 Optimization/Mathematical Programming Solvers

This solver category covers linear, integer, dynamic, quadratic, and network
models. The taxonomy of these with examples of corresponding solvers is provided
in Figure 3.
Results of a survey published in the June 2005 issue of OR/MS Today includes
over 40 software products divided into solvers, modeling environments, and linking
solvers. Notable examples are AIMMS, AMPL, Frontier Analyst, GAMS, ILOG
CPLEX, LAMPS, LINDO, LINGO, MOSEK, Premium Solver, SAS, Solver
Platform, TOMLAB, and Express.
 Model Management and Solvers for Decision Support 245

Figure 3. Sample taxonomy of optimization models and compatible solver assignment.

Source: Lee and Huh, 2003

5.1.2 Simulation solvers

Results of a simulation survey published in the December 2005 issue of OR/MS

Today includes the products, vendors, typical applications, and situations for which
the product fits. Notable products/vendors for this class of solvers include: C Risk
(Palisale Corp.), Agena Risk (Agena Inc.), Analytica (Lamina), Goldsim (GoldSim
Corp.), SDK (Frontline Systems), Simulab, VirtualSim, Volrerine, Vissim, and
Simprocess (CACI).

5.1.3 Decision Analysis Solvers

Last surveyed in December 2006, OR/MS Today includes many decision analysis
products with diverse functionalities. Representative products are Analytica,
Crystal Ball, Decision Explorer, DPL, Frontier Analyst, Insight, and Logical
Decisions.
246 Ting-Peng Liang et al.

5.1.4 Other Analytical Solvers

OR/MS Today reports on other solver categories from time to time. Examples are
vehicle-routing solvers (June 2006 issue) and forecasting solvers (August 2006
issue). Each survey is accompanied by an explanation and analysis.

5.2 Spreadsheet-Based Solvers: The Excel Family

A common platform for many commercial solvers is spreadsheets, primarily Excel,

via integrated algorithmic functions and add-on tools. Microsoft Excel Solver,
bundled with Excel and Microsoft Office, is the most widely used general-purpose
optimization modeling system. Because of the architecture of spreadsheet
programs, it is easy to create models that even contain discontinuous functions or
non-numeric values (these models usually cannot be solved with classical
optimization methods). Thus, solvers in a spreadsheet are often more useful than
modeling languages such as GAMS and AMPL [see Fylstra et al. (1998) for
details].
Excel-based solver tools are popular spreadsheet products that allow the solution
of a variety of mathematical programming optimization problems such as product
mix, capital budgeting, financial planning, workforce scheduling, and transpor-
tation/distribution problems. Users can enter an objective function, function value
(e. g., maximize monthly profit) in a “target cell,” parameters of constraints about
the information entered in “changing cells” (values that can be entered and changed
by the users), and “constraint,” which reflects the restrictions placed on the
“changing cells” values. For details, see: http://office.Microsoft.com/en-us/help/
HA011245961033.aspx.
In addition to Excel Solver for optimization, there are dozens of solvers for
statistical and mathematical calculation known as Excel Functions. These include
many “building blocks” that can be used within Excel’s spreadsheet applications. In
addition, there is a very large number of add-ins that expands the capabilities of
Excel. For example, What’s Best allows building of large-scale optimization
models in a free-form layout within a spreadsheet (http://www.lindo.com/products/
wb/wbm.html).
Many other products work with Excel Solver. For example, Solver.com offers
solvers/optimizers (and tutorials) for all the major spreadsheets (e. g., Excel, Lotus
1-2-3) using the Frontline systems. Its Premium Solver products work with Excel
Solver models, offer special features such as the Evolutionary Solver, and can
handle problems of virtually unlimited size. The Solver SDK products make it easy
for C#, VB.NET, C++, Visual Basic, Java, and MATLAB programs to solve
optimization problems.
The availability of a large number of commercially available solvers often makes
it difficult to select an appropriate solver(s) for a particular model and to adequately
apply the solver(s) to the model to be solved. To overcome this difficulty, a mo-
del-solver integration framework may be necessary. For this purpose, Lee and Huh
 Model Management and Solvers for Decision Support 247

(2003) suggest a methodology, based on Web Services to enable a DSS to

autonomously suggest compatible solver(s) and apply them to specific scenarios.
This was done by encapsulating individual services as Web Services, one of the
most promising future developments in model management, which is described in
the following section.

6 Impact of the Web Environment on Model

Management and Solvers

During the past decade, the World Wide Web has changed the design, development,
and implementation of information systems. It also has enabled a new way of
thinking about systematic approaches to decision making by allowing a distributed
implementation of decision support technologies (Bhargava et al., 2007; Ching
et al., 1991).

6.1 Web-Based Decision Support Systems

A Web-based decision support system is a computerized system that delivers

decision support information or decision support tools to a manager or business
analyst using a Web browser (Holsapple et al., 2000). The computer server that is
hosting the DSS application is linked to the user’s computer by a network with the
TCP/IP protocol. Web-based DSS may be web-enabled or web-based. The dif-
ference between web-enabled and web-based DSS is that the former may be
a traditional DSS that is compatible with the web environment for platform-in-
dependent information delivery to remote sites, while the latter has a technological
architecture designed to take full advantage of the distributed nature of the Internet
(Bhargava and Krishnan, 1998). Web-enabled systems provide a transitional
architecture between traditional DSS and totally web-based DSS.
Web-based decision support architectures can be classified into three types in
terms of technologies that enable the following (Bhargava et al., 2007):
• Server-side computation. The server provides most computing power to
facilitate platform-independence and universal access to decision support
applications. The client serves as a window between the user and the
system.
• Client-side computation. The server provides coordination while most
computation is done on the client computers. This allows higher local
control and more capabilities to be embedded in the user interface
including client-side scripting languages.
• A distributed implementation and deployment of DSS component. Both
servers and clients play roles in system execution.
248 Ting-Peng Liang et al.

From the coordination standpoint, there are two alternatives: centralized or

distributed. In centralized coordination, a single web server functions as a central
resource administrator that manages the various decision models, solvers, and data
that can be accessed via a web browser. For example, the NEOS server provides
a collection of optimization algorithms and develops an effective environment for
solving optimization problems over the Internet. Users can use a web browser to
submit their optimization problems. Once the optimization problem is submitted to
the NEOS Server, it is processed and dispatched for execution at a remote solver.
The NEOS Server maintains communication with the remote solver and returns
results to the user (Dolan et al., 2002; http://neos.mcs.anl.gov).
In the distributed approach, decision support resources are managed inde-
pendently from each other on the Internet. A DSS combines components from
various sources to deliver solutions to specific applications. For example, Dolk
(2000) proposes an integrated modeling environment as a distributed, compo-
nent-based, warehouse-driven software system to provide decision-making infor-
mation. Kim (2001) has developed an XML-based modeling language for the open
interchange of decision models. This language allows applications and OLAP tools
access to models obtained from various sources without having to handle individual
differences between MSOR and DSS communities.

6.2 Distributed Model Management

In a web-based DSS, decision models and data resources are distributed at different
locations with different platforms that are interconnected by computer networks. It
is therefore natural for researchers to explore how model management can be done
on the web. By distributed model management, we mean managing models in
a distributed environment (Huh et al., 1999).
For the MMS issues identified in Section 4, the major concern is how to link
component models and data seamlessly on the Internet. For model composition and
integration, components from different sources may create a number of problems
that can range from syntax compatibility to security control. In order to solve the
problem of model composition when models and data are distributed across
multiple sites, Chari (2003) extended the filter space approach (Chari, 2002) to
facilitate model composition when data resources are distributed at different sites
with overlapping key values and different scopes. Huh and Kim (2004) also
proposed a collaborative model management environment.
Model-data integration is another critical issue in a distributed environment.
Early research on model-data integration focused on how they could be integrated
seamlessly (e. g., Liang, 1985). Subsequent studies investigate actual applications
(Rizzoli et al., 1998) and integration with a data warehouse in distributed modeling
(Dolk, 2000). The interoperability in the distributed environment opens another
important issue: information security (Song et al., 2005).
 Model Management and Solvers for Decision Support 249

6.3 Enterprise Model Management

The distribution of decision models can be further extended to enterprise-level

decision support or enterprise modeling. Enterprise modeling (EM) is a set of
activities, processes, representations, and conceptualizations used to construct
models of the structure, behavior, and organization of the enterprise (Chapurlat
et al., 2006). Goul and Corral (2007) argue that enterprise model management
(EMM) is an important part of the next generation of decision support.
The purpose of EMM is to provide integrated decision support for enterprise.
Integrated decision support includes not only the personnel, processes, and systems,
but also the processes and systems used to coordinate distributed organizational and
inter-organizational requirements. Therefore, an integrated decision support
capability should accommodate different views of an enterprise including advances
in data warehouses, knowledge management, and model management. Research
topics for EMM identified by Goul and Corral (2007) include the following:

• Furthering the objectives of enterprise modeling, artifact and ontology

transformation, and mapping to unified representations.
• Defining and expanding the models and operators needed to manipulate,
integrate, maintain, and store those unified representations in facilitation
of organizational decision support services.
• Extending the notion of interoperability to higher levels of individual,
group, and organizational cooperation, collaboration, and inter-work
(i. e., inter-organizational work systems).
• Directing the design, development, and empirical study of efficient and
effective solutions for enabling these new, higher level notions of in-
teroperability.
• Advancing the role of integrated decision support in sustaining com-
petitive advantage for “networked” or “smart” organizations where inter-
works are common.
• Exploring new theories and expanding existing theories of inter-or-
ganizational decision support to advance knowledge about EMM repre-
sentations, operations, services, support for inter-works, and the capability
of integrated decision support to provide sustained competitive advantage.

6.4 Web Services Technology for Model Management

Another future research direction is the application of Web Services to model

management. Web Services technology is currently a popular technology for
distributed computing. Web Services are “loosely coupled, reusable software com-
ponents that semantically encapsulate discrete functionality and are distributed and
programmatically accessible over standard Internet protocols” (Gottschalk et al.,
2002; Ferris and Farrell, 2003). Figure 4 illustrates the conceptual architecture of
Web Services.
250 Ting-Peng Liang et al.

Figure 4. Conceptual architecture for Web Services

The benefit of using Web Services is twofold. First, in the technical dimension,
Web Services represent a collection of standard protocols for the creation,
distribution, discovery, and integration of semantic software components that
encapsulate business functionalities. Second, in the business dimension, Web
Services enable just-in-time software service provisioning through the integration
of loosely coupled software components.
Instead of viewing models as data or objects, Web Services leads to models
being viewed as services. Models are represented and implemented as computer
software according to the Web Services standard. The decision makers can invoke
decision support services and combine components from various sources at
run-time to find specific computational support.
Kwon (2003) suggests that Web Services technology is useful for at least two
reasons. First, Web Services do not need to know who or what will use the services
being provided. This frees the DSS developer from the burden of user interface.
Secondly, cross-platform capabilities necessary for inter-operability in a distributed
environment are much better for Web Services than a traditional technology in
a heterogeneous environment.
Several intentions of using Web Services for model management have been
reported. For instance, Iyer et al. (2005) propose a Web Service-based approach to
model management. They identify the layers of modeling knowledge and describe
an environment based on Web Services architecture that would help store, retrieve,
and distribute the layers of modeling language.
Güntzer et al. (2007) propose a new retrieval approach for Web Services that is
based on a structured modeling approach called Structured Services Models (SSM).
This approach can help users retrieve online decision support services.
 Model Management and Solvers for Decision Support 251

Mitra and Valente (2007) examine trends in the provision of optimization tools
and optimization-based decision support systems as remote applications. They
analyze the evolution from the Application Service Provision (ASP) model to the
e-Services model, and illustrate the importance of distributed optimization
components in the effective deployment of analytic applications in businesses. Web
Service technology is used to deliver optimization-based applications and
optimization components in a distributed environment.

6.5 Agent-Based Model Management

In a distributed environment, intelligent agent technologies offer an attractive

approach for handling complex tasks. An intelligent agent (IA) is computer
software that can perform certain tasks automatically to reduce the load of the
decision maker. IAs have been used widely on the Internet for electronic commerce
applications (e. g., Liang and Huang, 2000). In a distributed modeling environment,
intelligent agents can be used for handling certain modeling tasks. For instance, Liu
et al. (1990) propose an agent-based approach for model management. Lee and Huh
(2003) propose a conceptual model of using Web services for model-to-server
integration (see Figure 5). In their framework, model agents and solver agents work
closely together to provide necessary support.

Figure 5. Conceptual architecture of a model-solver integration framework. Source: Lee and

Huh (2003)
252 Ting-Peng Liang et al.

7 Conclusion

The increased interest in business analytics and the increased capabilities of

computers and web-based systems create an opportunity to provide users with DSSs
that facilitate better and faster decision making. Many of these DSSs need
computer-based decision models to help solve complex problems. These models
need to be properly managed by model management systems.
In this chapter we have presented the essentials of model management systems
and the process and tools of its implementation. While the computing environments
have been changed since the major research efforts of MMS took place in
1980–1998, the underlying issues remain the same. However, the new computing
capabilities and the Web environment provide opportunities for a more effective
and efficient utilization of MMS tools and solvers, and for decision support to
more-complex and less-structured problems.

References

Banerjee, S. and A. Basu, “Model Type Selection in an Integrated DSS

Environment,” Decis Support Syst, 9, 1993, 75−89.
Basu, A. and R.W. Blanning, “Metagraphs: A Tool for Modeling Decision Support
Systems,” Manage Sci, 40(12), 1994, 1579−1600.
Basu, A. and R.W. Blanning, “Model Integration Using Metagraphs,” Inf Syst Res,
5(3), 1994, 195−218.
Basu, A. and R.W. Blanning, “The Analysis of Assumptions in Model Bases Using
Metagraphs,” Manage Sci, 44(7), 1998, 982−995.
Bhargava, H.K. and R. Krishnan, “Computer-aided Model Construction,” Decis
Support Syst, 9(1), 1993, 91−111.
Bhargava, H.K., R. Krishnan and R. Muller, “Decision Support on Demand:
Emerging Electronic Markets for Decision Technologies,” Decis Support Syst,
19, 1997, 193−214.
Bhargava, H.K., S. Kimbrough and R. Krishnan, “Unique Names Violations:
A Problem for Model Integration,” ORSA J Comput, 3(2), 1991, 107−120.
Bhargava, H.K. and S.O. Kimbrough, “Model Management: An Embedded
Languages Approach,” Decis Support Syst, 10, 1993, 277−299.
Bhargava, H.K. and R. Krishnan, “The World Wide Web: Opportunities for
Operations Research and Management Science,” INFORMS J Comput, 10, 1998,
359–383.
 Model Management and Solvers for Decision Support 253

Bhargava, H.K., D.J. Power and D. Sun, “Progress in Web-based Decision Support
Technologies,” Decis Support Syst (2007, in press).
Binbasioglu, M. and M. Jarke, “Domain Specific DSS Tools for Knowledge-based
Model Building,” Decis Support Syst, 2, 1986, 213−223.
Blanning, R., “An Entity Relationship Approach to Model Management,” Decis
Support Syst, 2, 1986, 65−72.
Blanning, R., “A Relational Framework for Join Implementation in Model
Management Systems,” Decis Support Syst, 1, 1985, 69−81.
Blanning, R., “A Relational Framework for Model Management,” DSS-82
Transact, 1982, 16−28.
Blanning, R., “Model Management Systems: An Overview,” Decis Support Syst, 9,
1993, 9−18.
Blanning, R., A. Chang, V. Dahr, C. Holsapple, M. Jarke, S. Kimbrough, J. Lerch,
M. Prietula and A. Whinston, “Model Management Systems,” in Stohr, E. and
Konsynski, B. (eds.), Information Systems and Decision Processes. Los
Alamitos, CA: IEEE Computer Society Press, 1992.
Bonczek, R., C. Holsapple and A. Whinston, “Data Base Management Techniques
for Mathematical Programming,” Proceedings of the SIGMAP Bicentennial
Conference on Mathematical Programming, Washington, D.C., November
1976.
Bonczek, R., C. Holsapple and A. Whinston, “Model Management via Problem
Reduction,” Proceedings of the Hawaiian International Conference on Systems
Sciences, Honolulu, January, 1979.
Bonczek, R., C. Holsapple and A. Whinston, “The Evolving Roles of Models
within Decision Support Systems,” Decis Sci, 11, 1980, 337−356.
Bonczek, R., C. Holsapple and A. Whinston, Foundations of Decision Support
Systems. New York, NY: Academic Press, 1981a.
Bonczek, R., C. Holsapple and A. Whinston, “A Generalized Decision Support
System Using Predicate Calculus and Network Data Base Management,” Oper
Res, 29(2), 1981b, 263−281.
Bonczek, R., C. Holsapple and A. Whinston, “The Evolution from MIS to DSS:
Extension of Data Management to Model Management,” in Ginzburg, M.,
Reitman, W., Stohr, E. (eds.), Decision Support Systems. Amsterdam:
North-Holland, 1982.
Bonczek, R., C. Holsapple and A. Whinston, “Specifications of Modeling
Knowledge in Decision Support Systems,” in Sol, H. (ed.) Processes and Tools
for Decision Support. Amsterdam: North-Holland, 1983.
Bradley, G. and R. Clemence, “A Type Calculus for Executable Modeling
Languages,” IMA J Math Manage, 1(4), 1987, 277−291.
254 Ting-Peng Liang et al.

Chang, A., C. Holsapple and A. Whinston, “Model Management Issues and

Directions,” Decis Support Syst, 9(1), 19−37.
Chapurlat, V., B. Kamsu-Foguem and F. Prunet, “A Formal Verification
Framework and Associated Tools for Enterprise Modeling: Application to
UEML,” Comput Ind, 57, 2006, 153−166.
Chari, K., “Model Composition Using Filter Spaces,” Inf Syst Res, 13(1), 2002,
15−35.
Chari, K., “Model Composition in a Distributed Environment,” Decis Support Syst,
35(3), 2003, 399−413.
Chari, K. and Sen, T.K., “An Integrated Modeling System for Structured Modeling
Using Model Graphs,” INFORMS J Comput, 9(4), 1997, 397−416.
Ching, C., C. Holsapple and A. Whinston, “Computer Support in Distributed
Decision Environments,” in Sol, H. and Vecsenyi, J. (eds.), Environments for
Supporting Decision Processes. Amsterdam: North-Holland, 1991.
Chung, Q.B., T.R. Willemain and R.M. O’Keefe, “Influence of Model Management
Systems on Decision Making: Empirical Evidence and Implications,” J Op Res,
51(8), 2000, 936−948.
Dolan, E.D., R. Fourer, J.J. Moré and T.S. Munson, “Optimization on the NEOS
Server,” SIAM News, 35(6), 2002.
Dolk, D.R., “Integrated Model Management in the Data Warehouse Era,” Eur J Op
Res, 122, 2000, 199−218.
Dolk, D.K. and B.R. Konsynski, “Knowledge Representation for Model
Management Systems” IEEE T Soft Eng, SE-10, No. 6, 1984, 619−628.
Dolk, D.K. and J.E. Kottemann, “Model Integration and a Theory of Models,”
Decis Support Syst, 9, 1993, 51−63.
Dutta, A. and A. Basu, “Artificial Intelligence Approach to Model Management in
Decision Support Systems,” Computer, 17, 1984, 89−97.
Elam, J., “Model Management Systems: A Framework for Development,”
Proceedings of the 1980 Southwest AIDS Conference, Atlanta, GA, 1980.
Elam, J.J. and Konsynski, B.R., “Using Artificial Intelligence to Enhance the
Capabilities of Model Management Systems,” Decis Sci, 18(3), 1987, 487−502.
Ferris, C. and J. Farrell, “What are Web Services?” Comm ACM, 46(6), 2003, 31.
Foster, M.R., “Key Concepts in Model Selection: Performance and Generali-
zability,” J Math Psych, 44, 2000, 205−231.
Gagliardi, M. and C. Spera, “BLOOMS: A Prototype Modeling Language with
Object Oriented Features,” Decis Support Syst, 19, 1997, 1−21.
 Model Management and Solvers for Decision Support 255

Gagliardi, M. and C. Spera, “Toward a Formal Theory of Model Integration,”

Annals Oper Res, 58, 1995, 405−440.
Geoffrion, A.M., “An Introduction to Structured Modeling,” Manage Sci, 1987,
547−588.
Geoffrion, A.M., “Reusing Structured Models via Model Integration,” Proceedings
of 22nd Annual Hawaii International Conference on the System Sciences, IEEE
Computer Society, 1989, 601−611.
Geoffrion, A.M., “The SML Language for Structured Modeling: levels 1 and 2,”
Oper Res, 40(1), 1992a, 38−57.
Geoffrion, A.M., “The SML Language for Structured Modeling: levels 3 and 4,”
Oper Res, 40(1), 1992b, 58−75.
Gottschalk, K., S. Graham, H. Kreger and J. Snell, “Introduction to Web Services
Architecture,” IBM Sys J, 41(2), 2002, 170−177.
Goul, M. and K. Corral, “Enterprise Model Management and Next Generation
Decision Support,” Decis Support Syst, 43(3), 2007, 915−932.
Gray, P. Guide to IFPS (Interactive Financial Planning System). New York:
McGraw-Hill Book Company, 1987.
Greenberg, H. and F. Murphy, “Views of Mathematical Programming Models and
Their Instances,” Decis Support Syst, 13(1), 1995, 3−34.
Güntzer, U., R. Müller, S. Müller, and R.D. Schimkat, “Retrieval for Decision
Support Resources by Structured Models,” Decis Support Syst, 43(4), 2007,
1117−1132.
Haseman, W., C. Holsapple and A. Whinston, “OR Data Base Interface – An
Application to Pollution Control,” Comput Oper Res, 3(1), 1976, 229−240.
Holsapple, C., K.D. Joshi and M. Singh, “Decision Support Applications in
Electronic Commerce,” in Shaw, M. et al. (eds.), Handbook on Electronic
Commerce. Berlin: Springer-Verlag, 2000.
Hong, S.N. and M.V. Mannino, “Formal Semantics of the Unified Modeling
Language LU,” Decis Support Syst, 13(3−4), 1995, 263−293.
Huh, S.Y. and Q.B. Chung, “A Model Management Framework for Heterogeneous
Algebraic Models – Object-oriented Data Base Management Systems
Approach,” OMEGA-Int J Manage Sci, 23(3), 1995, 235−256.
Huh, S.Y., H.M. Kim and Q.B. Chung, “Framework for Change Notification and
View Synchronization in Distributed Model Management Systems,”
OMEGA-Int J Manage Sci, 27(4), 1999, 431−443.
Huh, S.Y. and H.M. Kim, “A Real-time Synchronization Mechanism for
Collaborative Model Management,” Decis Support Syst, 37(3), 2004, 315−330.
Iyer, B., G. Shankaranarayanan and M.L. Lenard, “Model Management Decision
Environment: A Web Service Prototype for Spreadsheet Models,” Decis Support
Syst 40(2), 2005, 283−304.
256 Ting-Peng Liang et al.

Jeusfeld, M. and T.X. Bui, “Distributed Decision Support and Organizational

Connectivity: A Case Study,” Decis Support Syst, 19, 1997, 215−25.
Jones, C.V., “An Introduction to Graph-based Modeling Systems, Part I:
Overview,” ORSA J Comput, 2(2), 1990, 136−151.
Jones, C.V., “An Introduction to Graph-based Modeling Systems, Part II:
Graph-grammars and the Implementation,” ORSA J Comput, 3(3), 1991,
180−206.
Jones, C.V. and R. Krishnan, “A Visual Syntax Directed Environment for
Automated Model Development,” SUPA Working Paper 9−89, Carnegie Mellon
University, 1992.
Kim, H.D., “An XML-based Modeling Language for the Open Interchange of
Decision Models,” Decis Support Syst, 31, 2001, 429−441.
Krishnan, R., “A Logic Modeling Language for Model Construction,” Decis
Support Syst, 6, 1990, 123−152.
Krishnan, R., “PDM: A Knowledge-based Tool for Model Construction,” Decis
Support Syst, 7, 1991, 301−304.
Krishnan, R. and Chari, K., “Model Management: Survey, Future Directions and
a Bibliography,” Interact Transact ORMS, 3(1), 2000, 1−19.
Krishnan, R., P. Piela and A. Westerberg, “Reusing Mathematical Models in
ASCEND,” in Holsapple, C. and Whinston, A. (eds.) Recent Developments in
Decision Support Systems. Berlin: Springer-Verlag, 1993, 275−294.
Krishnan, R., X. Li and D. Steier, “Development of a Knowledge-based Model
Formulation System,” Comm ACM, 35(5), 1992, 138−146.
Kwon, O.B., “Meta Web Service: Building Web-based Open Decision Support
System Based on Web Services,” Expert Syst Appl, 24, 2003, 375−389.
Lenard, M.L., “Representing Models as Data,” J Manage Inf Syst, 2(4), 1986,
36−48.
Lenard, M.L., “An Object-oriented Approach to Model Management,” Decis
Support Syst, 9(1), 1993, 67−73.
Liang, T.P., “Development of a Knowledge-based Model Management System,”
Oper Res 36(6), 1988, 849−863.
Liang, T.P., “Reasoning for Automated Model Integration,” Appl Artificial Int,
4(4), 1990, 337−358.
Liang, T.P., “Modeling by Analogy: A Case-based Approach to Automated Linear
Program Formulation,” in Proceedings of the 24th Hawaii International
Conference on System Sciences, Kuaui, Hawaii, 1991, 276−283.
 Model Management and Solvers for Decision Support 257

Liang, T.P. and J.S. Huang, “A Framework for Applying Intelligent Agents to
Support Electronic Commerce, Decis Support Syst, 28, 2000, 305−317.
Liang, T.P. and C.V. Jones, “Meta-design Considerations in Developing Model
Management Systems,” Decis Sci, 19, 1988, 72−92.
Liang, T.P. and B.R., Konsynski, “Modeling by Analogy: Use of Analogical
Reasoning in Model Management Systems,” Decis Support Syst, 9, 1993,
113−125.
Liu, J.I., D.Y.Y. Yun and G. Klein, “An Agent for Intelligent Model Management,”
J Manage Inf Sys, 7(1), 1990, 101−122.
Ma, P., F. Murphy and E. Stohr, “A Graphics Interface for Linear Programming,”
Comm ACM, 32(8), 1989, 996−1012.
Ma, P., F. Murphy and E. Stohr, “An Implementation of LPFORM,” INFORMS J
Comput, 8(4), 1996, 383−401.
Mannino, M.V., B.S. Greenberg and S.N. Hong, “Model Libraries: Knowledge
Representation and Reasoning,” ORSA J Comput, 2, 1990, 287−301.
Mitra, G., P. Valente, “The Evolution of Web-based Optimization from ASP to
E-services,” Decis Support Syst, 43(4), 2007, 1096−1116.
Muhanna, W., “On the Organization of Large Shared Model Bases,” Annals of
Oper Res, 38, 1993, 359−398.
Muhanna, W. and R.A. Pick, “Meta-Modeling Concepts and Tools for Model
Management: A Systems Approach,” Manage Sci, 40(9), 1994, 1093−1123.
Murphy, F. and E. Stohr, “An Intelligent System for Formulating Linear
Programs,” Decis Support Syst, 3, 1986, 39−47.
Myung, I.J., “The Importance of Complexity in Model Selection,” J Math Psych,
44, 2000, 190−204.
Pillutla, S.N. and B.N. Nag., “Object-oriented Model Construction in Production
Scheduling Decisions,” Decis Support Syst, 18(3−4), 1996, 357−375.
Power, D.J., ”Web-based Decision Support System,” DSstar, The On-Line Exec J
Data-Intens Decis Supp, 2(33, 34), 1998.
Raghunathan, S., R. Krishnan and J. May, “On Using Belief Maintenance Systems
to Assist Mathematical Modeling,” IEEE T Syst Man Cy, 25(2), 1995, 287−303.
Raghunathan, S., R. Krishnan and J. May, “MODFORM: A Knowledge-based Tool
to Support the Modeling Process,” Inf Syst Res, 4(4), 1993, 331−358.
Rizzoli, A.E., J.R. Davis and D.J. Abel, “Model and Data Integration and Re-use in
Environmental Decision Support Systems,” Decis Support Syst, 24(2), 1998,
127−144.
 258 Ting-Peng Liang et al.

Song, G.L., K. Zhang, L., Thuraisingham and J. Kong, “Secure Model Management
Operations for the Web,” Proceedings of Data and Applications Security XIX,
Lect Notes Comput Sci, 3654, 2005, 237−251.
Tseng, S.F., “Diverse Reasoning in Automated Model Formulation,” Decis Support
Syst, 20, 357−383, 1997.
Turban, E, J.E. Aronson, T.P. Liang and R. Sharda, Decis Supp Bus Intel Syst,
Upper Saddle River, NJ: Prentice-Hall, 2007.

View publication stats