Decision Sciences

Volume 30 Number 3
Summer 1999
Printed in the U.S.A.

The Conference Paper-Reviewer Assignment

Problem”
David Hartvigsen and Jerry C. Wei
D e p a m e n t of Management, College of Business Administration, University of Notre Dame,
Notre Dame, IN 46556-0399, email: hartvigsen. 1 C%hd.edu and [email protected]

Richard Czuchlewski
Department of Operations Researh and Industrial Engineering, Cornell University, Ithaca,
NY 14853

ABSTRACT
Conference organizers often face the following problem: Given a collection of submit-
ted papers, select a subset to be presented at the conference. The bulk of the work often
amounts to assembling a pool of reviewers and then sending each submitted paper to
several reviewers. We present in this paper a technique for finding a good assignment of
papers to reviewers. An important feature of the solution we find is that each paper is
sent to at least one reviewer who is “as expert as possible” for that paper. A major com-
ponent of the problem is modeled as a bottleneck version of a capacitated transshipment
problem.
Subject Areas: Assignment Problem, Network Theory, Optimization, and Ser-
vice Operations.

INTRODUCTION
Academic conferences are often organized in the following way. Several months
before the conference, an announcement is disseminated in which researchers are
asked to submit a written version of a paper they wish to present at the conference.
Larger conferences may have several “tracks,” in which case papers are earmarked
for the most appropriate track. Each (track) organizer next has the job of deciding
which papers to accept for presentation and which to reject. In some conferences
the organizer enlists the help of reviewers for this screening process. In this case
the organizer obtains a list of willing reviewers and then sends each paper to one
or more of them. The reviews that are sent back become the basis for the screening
process.

*The authors are grateful to Professors Patrick Shannon and Tom Foster of the College of Business and
Economics, Boise State University, Boise, Idaho. They were the co-chairpersons of the POM-Manufacturing
track of the 1998 Decision Sciences Institute annual meeting and provided the data of papers and reviewers
that were tested in the model.

865
 15405915, 1999, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5915.1999.tb00910.x by HUT - Hanoi Univ of Technology, Wiley Online Library on [24/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
866 The Conference Paper-Reviewer Assignment Problem

In this paper we consider the following problem of the organizer: How to best
assign the papers to the reviewers? Of course, the loose objective of the organizer is
to send each paper to reviewers who have expertise in the subject matter of the
paper. This goal can be difficult to achieve when (1) there are a large number of
papers and reviewers; (2) there is a wide variety of different subject areas among the
papers; and (3) there is a wide variety of types of expertise among the reviewers.
A standard method for solving this problem is to use some sort of classifica-
tion scheme for the papers and reviewers. For example, each author and reviewer
may be asked to classify his or her subject or expertise by selecting a few keywords
from a list supplied by the organizer. The organizer can then attempt to assign
papers to reviewers with common keywords. This might be done in a heuristic
fashion by simply considering the papers in some order and assigning each to any
underutilized reviewer with a common keyword. This can lead to two problems.
1. This heuristic does not guarantee an optimal solution, even in the sense
of‘ maximizing the number of assignments with keyword overlap. Even
an optimal solution, which maximizes the total number of overlaps, may
assign some papers to reviewers with no keyword overlap.
2. This heuristic does not address the following more subtle issue: The
method of overlapping keywords is fairly crude. To see this, note that
some pairs of keywords may be more similar to each other than other
pairs. Hence, two lists of keywords with no overlap may represent a
paper and reviewer that are completely dissimilar or are similar enough
for a reasonable assignment to be made.
In this paper we present an optimization approach to solving the paper-
reviewer assignment problem faced by a conference organizer. Our solution has
two phases. In the first phase, we describe a way of quantifying the level of exper-
tise of each reviewer for each paper. This method is more refined than counting
keyword overlaps and provides a soiution to Problem 2 mentioned above. This
method involves solving a number of small transportation problems. The second
phase produces an assignment using the levels of expertise from the first phase. It
is not a conventional assignment problem (or capacitated transshipment problem).
Instead we find an overall best assignment that also guarantees that every paper is
assigned, for example, to at least one reviewer who is “as expert as possible” for
that paper. This technique may be succinctly described as a bottleneck version of
the capacitated transshipment problem. This phase addresses Problem 1 above.
The remainder of the paper is organized as follows. The next section contains
a brief review of related work in the literature. The third section describes our
model and solution technique. The following section describes the implementation
of this method for the 1998 Decision Sciences Institute annual meeting. Then, gen-
eralizations and other possible applications of the model are discussed, followed
by a conclusion.

ASSIGNMENT PROBLEMS IN THE LITERATURE

The general assignment problem involves assigning members of one group of
objects (e.g., workers) to members of another group of objects (e.g., jobs). It is one
 15405915, 1999, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5915.1999.tb00910.x by HUT - Hanoi Univ of Technology, Wiley Online Library on [24/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Hartvigsen, Wei, and Czuchlewski 867

of the best known and most studied special cases of the minimum cost network
flow problem. The general assignment problem can also be viewed as a special
case of a transportation problem in which all the supplies and demands equal 1.
Ahuja, Magnanti, and Orlin (1993) provided an excellent review of solution meth-
ods and applications of the assignment problem. Among the applications they listed
are the following: personnel assignment (Machol, 1970; Ewashko & Dudding,
1971); optimal depletion of inventory (Derman & Klein, 1959); scheduling on par-
allel machines (Horn, 1973); pairing stereo speakers (Mason & Philpott, 1988);
and vehicle and crew scheduling (Carraresi & Gallo, 1984). Other applications are
posting military serviceman (Klingman & Phillips, 1984; Bausch, Brown, Hundley,
Rapp, & Rosenthal, 1991), airline commuting (Hansen & Wendell, 1982), and
classroom assignment (Carter & Tovey, 1992), among others.
The multiperiod assignment problem adds the complexity of the time dimen-
sion to the assignment decision. Many staff-planning and workforce-scheduling
applications can be solved as multiperiod assignment problems with various mod-
eling characteristics and solution approaches (Ross & Zoltners, 1979; Krajewski,
Ritzman, & McKenzie, 1980; Aronson, 1986; Mazzola & Neebe, 1986; Bechtold
& Showalter, 1987; Showalter & Mabert, 1988; Gilbert & Hofstra, 1988; Franz,
Baker, Leong, & Rakes, 1989; Franz & Miller, 1993). Most of the multiperiod
assignment formulations tend to use integer programming models and solve them
by heuristics that exploit the special structures of the models.

METHODOLOGY
We begin this section with a definition of our problem. We next describe our model
for this problem and our solution technique.
The problem we study is the following. We are given a set [ 1, . . ., r ) of
reviewers and a set { 1, . . ., p } of papers; we wish to assign the papers to the
reviewers so as to satisfy the following conditions.
(1.1) Each reviewer should be assigned at most three papers;
(1.2) Each paper should be assigned to exactly three reviewers;
(1.3) As much as possible, each paper should be assigned to reviewers who
are experts for that paper.
Of course, the number “three” in (1.1) and (1.2) is arbitrary and can be changed in
general (although we assume we have enough reviewers so that a feasible solution
exists). The first task in our solution method is the following.
Task 1: For each reviewer i and paper j , determine a number sij
denoting the “degree of expertise” of reviewer i for paper j .
The higher this number is, the better paperj falls within the
expertise of reviewer i.
We can now define the “weight of an assignment” to be the sum of the num-
bers sij for all pairs i j in the assignment. This suggests the idea of finding an assign-
ment by solving a maximum weight-capacitated transportation problem on the
following network:
 15405915, 1999, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5915.1999.tb00910.x by HUT - Hanoi Univ of Technology, Wiley Online Library on [24/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
868 The Conference Paper-Reviewer Assignment Problem

(1.4) For each reviewer there is a source node;

(1.5) For each paper there is a sink node;
(1.6) The supply at each source node is less than or equal to 3 units;
(1.7) The demand at each sink node is equal to 3 units;
(1.8) There is an arc from each source node i to each sink nodej with weight
sii’ capacity (or upper bound) 1, and lower bound 0.
Optimal solutions to this problem that are integral always exist, hence the
arcs at value 1 in such a solution define the assignment. However, a possible prob-
lem may result: A paper may be assigned to no reviewer with a “high degree” of
expertise for that paper. This can occur due to the “global” nature of the optimiza-
tion problem. We address this by providing procedures for the following two tasks:
Task 2: Given a threshold T, find a maximum weight assignment such
that every paper is assigned to at least one reviewer whose
expertise for that paper is greater khan or equal to T, or show
that no such assignment exists;
Task 3: Find the largest threshold F for which there exists a feasible
assignment in Task 2. This assignment is the solution to the
problem.
In other words, the two tasks above provide a solution of the following type.
First, find the highest threshold so that every paper can be sent to at least one
reviewer whose expertise is greater than or equal to the threshold. Then, subject to
this, an optimal solution also has the property that the sum of all the degrees of
expertise in the assignment is a maximum.
Let us observe that Tasks 2 and 3 are a generalization of the bottleneck
assignment problem. The bottleneck assignment problem is essentially the prob-
lem obtained by replacing the “three” in conditions (1.1) and (1.2) above with
“one” (see Ahuja et al., 1993).
In the remainder, we provide procedures for performing Tasks 1,2, and 3.

Procedure for Task 1

We use a three-step procedure to accomplish Task 1. In the first step we allow each
reviewer to classify himself or herself, and each author to classify his or her paper.
To do this we consider a standard set of categories { 1, . . ., C) of research areas. In
particular, each reviewer is given, say, 10 points and is instructed to distribute
these over the categories so as to characterize his or her areas of expertise. Like-
wise, each author is given 10 points to distribute over the categories so as to char-
acterize his or her paper. So, a reviewer may assign all 10 points to one category if
this is his or her sole area of expertise, or may spread them out over a number of
categories. The same holds for each author. Thus for each reviewer i we obtain a
reviewer classification vector:

R, = ( R i ( l ) ,..., R i ( C ) ) f o r i = 1,... r .
 15405915, 1999, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5915.1999.tb00910.x by HUT - Hanoi Univ of Technology, Wiley Online Library on [24/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Hartvigsen, Wei,and Czuchlewski 869

And for each paperj we obtain a paper classification vector:

Pi = ( P j ( l ) ,..., P j ( C ) )f o r j = 1, . . . p .

Note that

C C

k= 1 k= I

f o r i = 1,...r a n d j = 1,...p. (3)

In the second step we estimate numbers wkl,fork = 1 ,. . . ,C and 1 = 1 , . . . , C,

which denote the “degrees of similarity” between each pair of categories. A larger
number indicates a higher degree of similarity. These numbers can be obtained by
polling one or more experts and may represent a rounded average of their individ-
ual estimations.
In the third step, the degree of expertise numbers sij are found as solutions to
the following linear programs:

Each such linear program is a transportation problem from C sources

indexed by k to C sinks indexed by 1. Each source corresponds to a category, as
does each sink. The first group of constraints says that the supply at each source is
the number of points assigned to the corresponding category by reviewer i. The
second group of constraints says that the demand at each sink is the number of
points assigned to the corresponding category by the author of paper j. In practice,
this transportationproblem may be reduced to a transportation problem with fewer
constraints and variables, where the sources are those categories given a positive
value by reviewer i and the sinks are those categories given a positive value by the
author of paperj. A variable yk[ will tend to be large in an optimal solution if the
 15405915, 1999, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5915.1999.tb00910.x by HUT - Hanoi Univ of Technology, Wiley Online Library on [24/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
870 The Conference Paper-Reviewer Assignment Problem

following conditions hold: the reviewer assigned a significant number of points to

category k; the author assigned a significant number of points to category I ; and
categories k and 1 are similar.
The numbers sU so obtained have some nice properties. In particular, suppose
R,(k) = Pi(k)f o r k = 1 , . . ., C;that is, reviewer i and paperj are described by iden-
tical classification vectors. Then, it’s easy to see that sU takes on the maximum pos-
sible value. (If the degrees of similarity are drawn from (0, . . ., 5 ) and reviewers
and authors are allotted 10 points, then this maximum value is 50.) On the other
hand, if the degree of similarity between every category chosen by a reviewer and
an author is 0, then si, = 0. All numbers between these extremes are possible;
hence, we obtain a more subtle measure of the degree of expertise of a reviewer for
a paper than keyword overlaps (as discussed in the Introduction).

Procedure €or Tasks 2 and 3

Task 2 is performed by adding more structure to the network described in (1.4)-
( 1.8) and, thus, transforming a maximum weight capacitated transportation prob-
lem into a maximum weight capacitated transshipment problem. We construct the
following network N with supplies and demands. Recall that T is a threshold value
given in Task 2 (see Figure 1 f o r an example of such a network).
1. For each reviewer i there is a source node si.
2. For each paper j there are two nodes tj and t J,where 5 is a sink node and
t J is an intermediate node.

3. The supply at each source node is less than or equal to 3 units.

4. The demand at each sink node is equal to 3 units.
5. There is an arc from node sito 5 iff sU < 7\, such an arc has weight sU and
capacity 1.
6 . There is an arc from node sito t3 iff sii 2 T; such an arc has weight sU and
capacity 1.
7 . For every paperj, there is an arc from t j to 5 with weight 0 and lower
bound 1.
8. For every paper j , there is an arc from 5 to t j with weight -M, for some
large positive number M .
9. Unless otherwise stated above, all arcs have a lower bound of 0 and a
capacity (or upper bound) of +-.

Procedure for Task 2

Step 1: Form the network N described above.
Step 2: Find a maximum weight transshipment solution for N.If the solu-
tion is positive, then output the assignment found (as indicated by
the arcs with one unit of flow); otherwise, the original problem in
Task 2 has no feasible solution.
 15405915, 1999, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5915.1999.tb00910.x by HUT - Hanoi Univ of Technology, Wiley Online Library on [24/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Hartvigsen, Wei, and Czuchlewski 87 1

Figure 1: Network model.

reviewers papers
. I

tl

13 0=3

t._
13 0=3

t3

13 0=3

For an excellent treatment of transshipment problems (also called minimum

cost flow problems), both mathematical formulations and solution techniques, the
reader is referred to Ahuja et al. (1993).

Procedure for Task 3

Step 1: Set T to be the smallest value so. Apply the Procedure for Task 2.
Step 2: Set T to be the next smallest value sii' Apply the Procedure for Task 2.
If a feasible assignment is found, repeat this step. If no feasible solu-
tion is found, output the assignment found in the previous iteration.
Observe that the use of intermediate nodes and the large negative weights -M
in the construction of N gives us the type of "bottleneck" solution that we want.
Also observe that the maximum weight capacitated transshipment problem
defined by the network N can be solved with any software for minimum weight
capacitated transshipment problems (or minimum cost flow problems) by simply
changing the signs of the weights (the resulting network has no negative weight-
directed cycles). Finally, observe that Step 2 of the Procedure for Task 3 can be
more efficiently implemented with a binary search (divide and conquer).

IMPLEMENTATION
In this section we describe a particular application of the methodology proposed in
the previous section. This application was for the 1998 annual meeting of the Deci-
sion Sciences Institute in Las Vegas. Annual meetings for this organization typi-
cally consist of around 1,000 paper submissions that are divided into 12 functional
 15405915, 1999, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5915.1999.tb00910.x by HUT - Hanoi Univ of Technology, Wiley Online Library on [24/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
872 The Conference Paper-Reviewer Assignment Problem

tracks. We applied our methodology to the Production and Operations Manage-

ment (P0M)-Manufacturing Track, which has always been one of the largest
tracks, with around 200 papers. A common practice for the track organizer is to
send each paper to three reviewers and to ensure that each reviewer receives at
most three papers.
The first task in applying our methodology was to construct a set of catego-
ries for the classification of papers and reviewers. We consolidated a list of 48 cat-
egories used by the Decision Sciences Journal for Operations and Logistics
Management into a list of 30 categories for this purpose. We then constructed two
survey forms, one for authors and one for reviewers. In March 1998, the POM-
Manufacturing Track received feedback from 182 volunteer reviewers who filled
out the survey form. The track organizers then filled out the paper survey after a
quick review of each of the 174 papers submitted to the track. (Alternatively, this
could have been done by the authors themselves.)
The next step was to construct a matrix that contains the “degrees of similar-
ity” between categories, that is, the numbers wk,defined in the previous section.
These are contained in a 30x30 “proximity” matrix. We used a scale of 0 to 5 ,
where 5 indicates the strongest possible similarity between two categories. This
proximity matrix provides one advantage over the traditional method of assigning
a paper to reviewer by keyword overlapping, as described in the Introduction. Uti-
lizing the proximity matrix, our methodology offers the possibility of finding
“almost expert” reviewers when “experts” are not available.
With 182 reviewers and 174 papers, we had to solve 31,668 transportation
problems to obtain the degree of expertise for each paper-reviewer pair. We
recorded the data in Microsoft Excel 97, wrote macros to organize data input and
output, and solved the transportation problems via macro calls to the Solver mod-
ule within Excel. We chose to use Excel due to its almost universal availability.
Although the number of transportation problems is huge, these problems are com-
pletely independent of each other and, hence, can be solved in parallel. We
acquired a computer lab and ran 18 Windows-based Pentium-133 personal com-
puters at the same time. Each computer (except the 18th one) took 30 minutes to
solve the transportation problems required.
We recorded all the 3 1,668 degrees of expertise numbers in a matrix. This
matrix yielded the weights on the arcs of the networks N that were constructed dur-
ing the running of the procedures for Tasks 2 and 3. We found that among the 182
reviewers, about 50 reviewers also submitted papers to the POM-Manufacturing
Track. To avoid sending a paper to its own author(s), we manually assigned a big
negative weight to all such combinations in the matrix. We then took this revised
matrix and a starting threshold T of 25 (as dictated by binary search), and gener-
ated (with an Excel macro) the network N in a form appropriate for input to the
SAS/OR software. (Again, we chose SAS/OR because of its wide availability.)
Each minimum cost capacitated transshipment problem had over 40,000 arcs, but
it only took the Windows version of SAS/OR about 10 seconds to solve each.
Using binary search, we quickly found that the largest threshold that admitted a
feasible solution was T = 30. This solution guaranteed that for each paper at least
one of the three reviewers had an expertise level of 30 or more and that this was not
possible for an expertise level of 3 1. The solution was also the best possible with
 15405915, 1999, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5915.1999.tb00910.x by HUT - Hanoi Univ of Technology, Wiley Online Library on [24/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Hartvigsen, Wei,and Czuchlewski 873

respect to the objective function subject to the threshold constraint. Because we

had more reviewers than papers, some reviewers were assigned fewer than three
papers in our solution. We also considered the problem of finding the largest
threshold such that every paper is assigned to two reviewers with expertise, which
is at least (as large as) this value. In this case, the largest threshold that admitted a
feasible solution was 24. This illustrates the trade-off between requiring one expert
reviewer for each paper versus requiring two expert reviewers. The three-reviewer
version of this problem also yielded a threshold of 24, although the objective func-
tion value dropped off slightly.

GENERALIZATIONS AND OTHER APPLICATIONS

The model we have presented is customized for a particular application. However,
many features of the model can be generalized and other applications are possible.
We first list below some of the possibilities for generalizing the model.
1. The number “three” in (1.1) and (1.2) can be any positive number and
need not be the same for all reviewers and papers. This will affect items
3 and 4 in the construction of N .
2. The type of inequality or equality required in (1.1) and (1.2) is not
important and need not be the same for all reviewers and papers. This
will affect items 3 and 4 in the construction of N .
3. In Task 2, the requirement “at least one” can be “at least 1” for some pos-
itive integer 1. This will change the lower bounds on the arcs in item 7 to
1 in the construction of N .
4. In Task 2, the requirement “every paper” can be replaced with “all but m
papers.” That is, an assignment may be considered feasible if no more
than m papers are assigned to reviewers who are not above the threshold
T. This may allow a higher threshold to be achieved for the vast majority
of the papers. This can be accomplished with the procedures as given
due to the use of the large number M . To see this, observe that in Step 2
of the Procedure for Task 2, a solution is always found to the transship-
ment problem, which minimizes the number of papers that are assigned
to reviewers whose level of expertise for that paper is less than the
threshold. Thus, the Procedure for Task 3 can be run until no feasible
solution is found in this generalized sense.
5. The use of 10 points in the classification vector is arbitrary.
6. The model described does not take into consideration the skill level of
the reviewers. This can be a function of experience, academic rank, or
reputation. However, an additive or multiplicative factor could be used
to adjust the levels of expertise sii appropriately. For example, suppose
we classify the reviewers into three types: studendassistant professor;
associate professor; full professor. We could assign weights, say 1,2,3,
to the respective types. Finally, we could replace each expertise level sjj
with d, sii’ where d is the weight for the type of reviewer i.
 15405915, 1999, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5915.1999.tb00910.x by HUT - Hanoi Univ of Technology, Wiley Online Library on [24/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
874 The Conference Paper-Reviewer Assignment Problem

Finally, we hope this methodology can be used in other applications that

require a good match between two groups. Some possible applications are given
below.

Consulting
Consider the situation of a number of projects and a number of people that must be
assigned to these projects. Each project requires a certain number of people and
each person must be assigned to a certain number of projects. Each person’s skills
can be characterized by distributing points over some categories, and each
project’s requirements can be characterized by distributing the same number of
points over the same categories. The expertise level of each person for each project
can be calculated with Task 1 of our model. The assignment of people to projects
can then be carried out using Task 2 of our model so that every project is assigned
at least one person who is as expert as possible.

Job interviews
Consider the problem of assigning graduating students to job interviews at a uni-
versity. Each company is allowed to interview a certain number of students and
each student is allowed to interview with a certain number of companies. To begin,
each student distributes a certain number of points over the interviewing compa-
nies according to their preferences, and each company distributes the same number
of points over the students according to their preferences. A straightforward ver-
sion of Task 1 of our model can be used to determine the strength of each student-
company pair. Task 2 of our model can then be used to assign students to inter-
viewers so that every student gets to interview with at least one company in which
they are strongly interested (this could alternatively be done from the companies’
points of view).

Class registration
Consider the situation of students requesting courses in a preregistration process at
a university. To begin, each student is given some number of points to distribute
over the courses offered (the more points they give to a course, the higher their
preference). The objective is to assign the students to courses so that each student
gets one (or two, etc.) course(s) with a point rating above a threshold that is made
as large as possible. Each course can also have a student capacity. (This model uses
only Task 2 of our model.)

CONCLUSIONS
In this paper we proposed a two-phase optimization approach to solving the con-
ference paper assignment problem. We implemented the proposed approach to
help organize the POM-Manufacturing Track of the 1998 annual meeting of the
Decision Sciences Institute. We provided optimal solutions to the assignment of
174 papers to 182 reviewers in this track. These solutions have the following prop-
erty: For each paper, a specified number (one, two, or three in our case) of review-
ers possess a level of expertise that is above a threshold, which is as large as
possible. [Received: May 20, 1998. Accepted: October 16, 1998.1
 15405915, 1999, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5915.1999.tb00910.x by HUT - Hanoi Univ of Technology, Wiley Online Library on [24/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Hartvigsen, Wei, and Czuchlewski 875

REFERENCES
Ahuja, R. K., Magnanti, T. L., & Orlin, J. B. (1993). Networkflows. Upper Saddle
Ridge, NJ: Prentice-Hall.
Aronson, J. E. (1986). The multiperiod assignment problem: A multicommonality
network flow model and specialized branch and bound algorithm. European
Journal of Operational Research, 23, 367-38 1,
Bausch, D. O., Brown, G. G., Hundley, D. R., Rapp, S. H., & Rosenthal, R. E.
(1991). Mobilizing marine corps officers. Interfaces, 21(4), 26-38.
Bechtold, S. E., & Showalter, M. J. (1987). A methodology for labor scheduling in
a service operating system. Decision Sciences, 18, 89- 107.
Carraresi, P., & Gallo, G. (1984). Network models for vehicle and crew schedul-
ing. European Journal of Operational Research, 16, 139-151.
Carter, M. W., & Tovey, C. A. (1992). When is the classroom assignment problem
hard? Operations Research, 40( l), S28-S39.
Derman, C., & Klein, M. (1959). A note on the optimal depletion of inventory.
Management Science, 5,210-214.
Ewashko, T. A., & Dudding, R. C. (1971). Application of Kuhn’s Hungarian
assignment algorithm to posting servicemen. Operations Research, 19,991.
Franz, L. S., & Miller, J. L. (1993). Scheduling medical residents to rotations:
Solving the large-scale multiperiod staff assignment problem. Operations
Research, 41(2), 269-279.
Franz, L. S., Baker, H. M., Leong, G. K., & Rakes, T. R. (1989). A mathematical
model for scheduling and staffing multiclinic health regions. European Jour-
nal of operational Research, 41,277-289.
Gilbert, K. C . , & Hofstra, R. B. (1988). Multidimensional assignment problems.
Decision Sciences, 19, 306-321.
Hansen, P., & Wendell, R. E. (1982). A note on airline commuting. Interfaces, 11
(12), 85-87.
Horn, W. A. (1973). Minimizing average flow time with parallel machines. Oper-
ations Research, 21,846-847.
Klingman D., & Phillips, N. (1984). Topological and computational aspects of pre-
emptive multicriteria military personnel assignment problems. Management
Sciences, 30( I), 1362-1375.
Krajewski, L. J., Ritzman, L. P., & McKenzie, J. P. (1980). Shifting scheduling in
banking operations: A case study application. lnterfaces, 10, 1-6.
Machol, R. E. (1970). An application of the assignment problem. Operations
Research, 18,745-746.
Mason, A. J., & Philpott, A. B. (1988). Pairing stereo speakers using matching
algorithms. Asia-Pacific Journal of Operational Research, 5, 101-116.
Mazzola, J. G., & Neebe, A. W. (1986). Resource-constrained assignment sched-
uling. Operations Research, 34,560-572.
 15405915, 1999, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5915.1999.tb00910.x by HUT - Hanoi Univ of Technology, Wiley Online Library on [24/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
876 The Conference Paper-Reviewer Assignment Problem

Ross, G. T., & Zoltners, A. A. (1979). Weighted assignment models and their
application. Management Science, 25,683-696.
Showalter, M. J., & Mabert, V. A. (1988). An evaluation of a full-/part-time tour
scheduling methodology. International Journal of Operations and Produc-
tion Management, 8, 54-7 1.

David Hartvigsen is an associate professor of management in the College of

Business Administration at the University of Notre Dame. He received his PhD in
mathematics from Carnegie-Mellon University in 1984. Some of the journals he
has published in are the SIAM Journal on Optimization, SIAM Journal on Discrete
Mathematics, ORSA Journal on Computing, and Mathematics of Operations
Research. He is a member of INFORMS, MPS, and A M S .

Jerry C. Wei is an associate professor of operations management at the University

of Notre Dame. He received a BS from National Tsing Hua University in Taiwan,
an ME from Rochester Institute of Technology, and a PhD in operations
management from Texas A&M University. He is on the editorial review board of
the Journal of Operations Management and has published in Decision Sciences,
European Journal of Operational Research, International Journal of Production
Research, Journal of Operational Research Society, Journal of Quality Technology,
and Journal of Manufacturing Systems. His current research interests include
supply chain management, master planning, J I T production, cellular
manufacturing, and manufacturing flexibility.

Richard Czuchlewski is currently working for the SABRE Group. He received an

MS in operations research from Cornell University in 1999 and a BS in
mathematics from the University of Notre Dame in 1998. He is a member of
INFORMS.