Hindawi

Scientific Programming
Volume 2017, Article ID 9057947, 7 pages
https://doi.org/10.1155/2017/9057947

Research Article
Mixed-Integer Linear Programming Models for
Teaching Assistant Assignment and Extensions

Xiaobo Qu,1 Wen Yi,2 Tingsong Wang,3 Shuaian Wang,4 Lin Xiao,5 and Zhiyuan Liu6
1
School of Civil and Environmental Engineering, University of Technology Sydney, Sydney, NSW 2007, Australia
2
Department of Building and Real Estate, The Hong Kong Polytechnic University, Kowloon, Hong Kong
3
School of Economics and Management, Wuhan University, Wuhan 430072, China
4
Department of Logistics & Maritime Studies, The Hong Kong Polytechnic University, Kowloon, Hong Kong
5
National Research Council of the National Research Academies of Science, Engineering, and Medicine, Washington, DC 20001, USA
6
Jiangsu Key Laboratory of Urban ITS, Jiangsu Province Collaborative Innovation Center of Modern Urban Traffic Technologies,
School of Transportation, Southeast University, Jiangsu, China

Correspondence should be addressed to Tingsong Wang; [email protected], Shuaian Wang; [email protected],

and Zhiyuan Liu; [email protected]

Received 7 May 2016; Revised 1 November 2016; Accepted 7 December 2016; Published 11 January 2017

Academic Editor: Christoph Kessler

Copyright © 2017 Xiaobo Qu et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

In this paper, we develop mixed-integer linear programming models for assigning the most appropriate teaching assistants to
the tutorials in a department. The objective is to maximize the number of tutorials that are taught by the most suitable teaching
assistants, accounting for the fact that different teaching assistants have different capabilities and each teaching assistant’s teaching
load cannot exceed a maximum value. Moreover, with optimization models, the teaching load allocation, a time-consuming process,
does not need to be carried out in a manual manner. We have further presented a number of extensions that capture more practical
considerations. Extensive numerical experiments show that the optimization models can be solved by an off-the-shelf solver and
used by departments in universities.

1. Introduction There are many factors to consider in TA assignment.

First, all tutorials must be taught in order to satisfy the
Teaching assistants (TAs) are essential in many departments needs of the teaching programs. Second, a TA should not be
in universities because of limited numbers of lecturers. TAs allocated a higher teaching load other than what is required
are mainly postgraduate students, though in some universi-
by her/his role because of her/his limited available time.
ties TAs can also be undergraduate students. TAs are usually
Third, TAs should teach tutorials that they are proficient with;
in charge of teaching tutorials in small classes. TA allocation,
otherwise it will cost them a lot of preparation time and
that is, which TA teaches which tutorial, is an essential task
that is carried out in a department every semester. It has been the teaching outcomes will not be ideal. Fourth, we should
well recognized that appropriate teaching staff allocation consider whether two tutorials are delivered at the same time,
contributes to the equality of teaching [1, 2]. The teaching and if so, they cannot be taught by the same TA. Fifth, it is easy
assistants are usually reallocated every semester. This is due to to see that if a TA teaches several tutorials of the same lecture,
the following two reasons: (1) changes of staff (some TAs are in other words, repeating the tutorials, then it is easier for the
no longer available because they have graduated or have other TA to save time for preparation. Sixth, it might be difficult for
commitments and new TAs are available) and (2) changes a lecturer to manage if there are too many TAs for the lecturer
of programs (some tutorials are cancelled and some new and hence it may be desirable to control the number of TAs
tutorials are developed). for a lecturer.
 5192, 2017, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2017/9057947 by Readcube (Labtiva Inc.), Wiley Online Library on [18/06/2024]. 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
2 Scientific Programming

Our personal experiences show that almost all depart- assistant should be in line with his or her availability, is
ments allocate teaching assistants manually in a trial-and- also very important. Burgess [12] did a comparison study
error manner. As we all know, this allocation is very time- regarding different methods for allocating teaching tasks to
consuming process and may lead to unfavorable results which faculty members. Vardi [13] carried out a similar work with
may jeopardize the teaching quality. In this regard, it is the objective of analyzing the impacts of workload allocation
imperative and of vital significance to develop a systematic on the satisfaction of working life of faculty members. Bentley
tool in order to better allocate the teaching assistants in a and Kyvik [14] compared the teaching workload allocation
more efficient manner. statuses in a number of countries. There are a number of
works related to the relations between research and teaching
1.1. Literature Review. It is universally acknowledged that for academic staff (e.g., [15–18]). It should be stressed that
improving the teaching quality is the number one con- all of the studies realized that equity has a vital impact on
cern in all teaching activities. Academics have developed academics’ satisfaction. Unfortunately, two issues are not
a number of new educational theories and pedagogical properly addressed. First, the equity issue is largely neglected
techniques to meet the teaching requirements in view of the in the above teaching load allocation optimization models.
changing society (e.g., [3, 4]). Some studies are related to Second, to the best of our knowledge, there is no research
the timetabling of educational organizations. Valouxis and which focuses on the equity issue for teaching assistants.
Housos [5] applied a constraint programming approach to In this study, we aim to develop teaching assistant
solve a high school timetabling problem considering various assignment models in order to optimize the teaching quality
practical constraints. Beligiannis et al. [6] proposed an by taking into account various constraints such as equality
adaptive algorithm to address the timetabling problem of an and availability of teaching assistants. The model can improve
educational institute. Pillay and Banzhaf [7] examined the use the “fitness” between teaching assistants and tutorials without
of genetic algorithms to address an examination timetabling jeopardizing the equity issue of teaching assistants. Further,
problem. A category of research that is more closely related this proposed model will reduce the workload for school
to ours is optimization models proposed to better allocate heads, managers, or program coordinators by automatically
teaching load for full time academic teaching staff in a selecting the optimal solutions.
teaching unit. For example, Breslaw [8] was among the first
to develop a linear programming model to address a teaching 1.2. Objectives and Contributions. The objective of this
staff assignment problem. His objective was to maximize research is to develop mixed-integer programming models
the preference of the faculty members. Schniederjans and [19–23] that are able to generate the optimal allocation
Kim [9] adopted a different mathematical approach—a goal plan of teaching assistants. The contribution of the paper
programming model—to optimize the overall benefit to the is threefold: First, we develop a new approach that is able
teaching department. Badri [10] developed a more complex to enhance the overall teaching performance or quality of
model that consists of a two-stage multiobjective scheduling a teaching unit essentially at no cost. This is due to the
approach. The objectives are twofold: first, the preference for fact that our model enables more teaching assistants to
tutorials by faculty members is incorporated; second, the time teach tutorials that they are competent at. As a result, the
slots of the tutorials that are preferred by faculty members are teaching assistants’ satisfaction can be improved and teaching
also considered. Qu et al. [11] developed a very simple model quality is accordingly improved. The basic assumption is
for assignment lecturers to classes and tested the model with very reasonable: the teaching quality is positively related to
data from a university in Australia. As discussed in Badri the “fitness” between teaching assistants and tutorials that
[10], the abovementioned mathematical approaches could they teach. However, it is very challenging if we want to
improve teaching quality by maximizing the preference of the reach the optimal solution manually: one cannot guarantee
faculty members. However, two key issues are neglected in that the optimal decision is made. By contrast, mathematical
the abovementioned studies. First, as the focus of teaching programming approaches could obtain optimal decisions for
activities is to improve or guarantee the teaching quality, department heads to assign the teaching tasks.
the teaching quality should be considered as the objective to Second, this allocation is usually done by senior aca-
maximize, rather than a constraint in the optimization model. demics (e.g., head of school, deputy head of school, depart-
In other words, teaching quality is much more important than ment chair, school manager, and program coordinator).
faculty preference as the former is the number one concern They are naturally very busy with their teaching, research,
in all teaching activities. Second, the abovementioned studies supervision, and other commitments. Our model can save the
are mainly focused on teaching load allocation for full time valuable time for these academics.
academic staff, while the arrangement for teaching assistants Third, the model considers the maximum teaching load
is mostly treated in a very simple way, which may lead to of each teaching assistant. Teaching assistants are usually on
unreliable results. a part-time basis. Imposing too much teaching to them will
In most teaching units, teaching assistants are Ph.D. or jeopardize their performance in their full time commitment
senior students and they are working as teaching assistants (i.e., study/research). If this allocation is carried out manually
on a part time basis. As a result, it is not appropriate to by senior academics, it is likely that the maximum teaching
assign heavy teaching loads to them, due to the fact that they is violated due to the fact that manual checking is very
need to focus on their own studies/research. In this regard, time-consuming. This deficiency can be completely overcome
the workload equity, namely, the workload of a teaching using our models.
 5192, 2017, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2017/9057947 by Readcube (Labtiva Inc.), Wiley Online Library on [18/06/2024]. 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
Scientific Programming 3

Fourth, we also consider whether two tutorials are deliv- the total number of tutorials delivered by the most appropri-
ered at the same time, and if so, they cannot be taught by the ate teaching assistant. At the same time, some other practical
same TA. factors must also be taken into account. For instance, we
Fifth, it is easy to see that if a TA teaches several tutorials should consider whether two tutorials are delivered at the
of the same lecture, in other words, repeating the tutorials, same time, and if so, they cannot be taught by the same TA;
then the TA can save time for preparation. Our model thus it is easy to see that if a TA teaches several tutorials of the
aims to increase the number of tutorial repetitions. same lecture, in other words, repeating the tutorials, then the
Sixth, it might be difficult for a lecturer to manage if TA can save time for preparation; it might be difficult for a
there are too many TAs for the lecturer and hence it may be lecturer to manage if there are too many TAs for the lecturer
desirable to control the number of TAs for a lecturer. This and hence it may be desirable to control the number of TAs
factor is also formulated in our model. for a lecturer.
The remainder of this paper is organized as follows.
Section 2 describes considerations in teaching assistant allo- 3. Basic Optimization Model
cation. Section 3 builds a basic integer linear optimization
model. Section 4 reports a number of practical extensions To address the teaching assistant (TA) assignment problem,
to the basic model. The results of numerical experiment are we develop an integer optimization model similar to Qu et
reported in Section 5. Conclusions are presented in Section 6. al. [11]. However, we will present rich extensions to the basic
optimization in the next section. The notations used in the
2. Problem Description basic optimization model are listed below.

This study examines the allocation of teaching assistants at Sets

a department or school. In a department, many tutorials
for undergraduates and postgraduates are taught by teaching 𝐼: set of tutorials
assistants. A teaching assistant has limited time that can be 𝐼𝑗 : set of tutorials that TA 𝑗 ∈ 𝐽 may teach; 𝐼𝑗 = {𝑖 ∈
spent on teaching. Different teaching assistants have different 𝐼 | 𝑗 ∈ 𝐽𝑖 }
maximum numbers of teaching hours because (i) they have
𝐽: set of TAs
different availability and (ii) they have different willingness
to teaching. In reality, most teaching assistants are Ph.D. 𝐽𝑖 : set of TAs who can deliver tutorial 𝑖 ∈ 𝐼; 𝐽𝑖 = 𝐽𝑖1 ∪ 𝐽𝑖2
students in a teaching unit. Some students have research
scholarships which limit the number of hours that they can 𝐽𝑖0 : set of TAs who cannot deliver tutorial 𝑖 ∈ 𝐼
spend in teaching, while the others have no scholarships so 𝐽𝑖1 : set of TAs who can teach but are not the most
that the number of hours that they can spend in teaching suitable for delivering tutorial 𝑖 ∈ 𝐼
is only limited by the visa requirement (for international
Ph.D. students). As different teaching assistants have different 𝐽𝑖2 : set of TAs who are the most suitable for delivering
expertise, we classify the relation between a tutorial and a tutorial 𝑖 ∈ 𝐼
teaching assistant into three types. (i) In the first type, the
teaching assistant is unable to teach that tutorial. For example,
Indices
in a department of mathematics, a teaching assistant with
expertise in statistics cannot teach the tutorial for “Group 𝑖: an index 𝑖 ∈ 𝐼 that refers to a particular tutorial
Theory.” (ii) In the second type, the teaching assistant can
teach a tutorial, but it is not the most suitable person. For 𝑗: an index 𝑗 ∈ 𝐽 that refers to a particular TA
example, a teaching assistant with expertise in the structural
engineering can teach the elective tutorial “Introduction to
Parameters
Geotechnical Engineering” for civil engineering students,
but she/he is not the most suitable person for teaching it, 𝑛𝑗min : the minimum number of tutorials that must be
as a teaching assistant with the expertise in geotechnical
delivered by TA 𝑗 ∈ 𝐽: it can be set at 0 if there is no
engineering is more suitable. (iii) In the third type, the
such requirement
teaching assistant is one of the most suitable persons to teach
a tutorial. For example, a teaching assistant with expertise in 𝑛𝑗max : the maximum number of tutorials that can be
transport engineering should teach the tutorial “Traffic Flow delivered by TA 𝑗 ∈ 𝐽
Theory” or “Traffic Engineering Fundamentals” if possible.
Another example is that if the teaching assistant taught this 𝑡𝑖 : the number of contact hours required for tutorial
tutorial and was highly appraised last year, she or he should 𝑖 ∈ 𝐼: in practice, 𝑡𝑖 is usually 1, 2, or 3 hours
also teach it this year if possible. 𝑡𝑗min : the minimum number of available hours per
We can now see that the purpose of optimizing the week for TA 𝑗 ∈ 𝐽
allocation of teaching assistants includes constraints that all
tutorials are taught and that the available teaching time of 𝑡𝑗max : the maximum number of available hours per
teaching assistants is satisfied and an objective of maximizing week for TA 𝑗 ∈ 𝐽
 5192, 2017, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2017/9057947 by Readcube (Labtiva Inc.), Wiley Online Library on [18/06/2024]. 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
4 Scientific Programming

Decision Variables research. TAs in one subarea may not be able to teach
tutorials in other subareas. As a result, the overall model for a
𝑥𝑖𝑗 : a binary decision variable which equals 1 if TA 𝑗 ∈ department can be decomposed for each area or subarea. Our
𝐽𝑖1 ∪ 𝐽𝑖2 delivers tutorial 𝑖 ∈ 𝐼 and 0 otherwise numerical experiments in Section 5 show that the model can
be solved by off-the-shelf solvers such as CPLEX.
The above sets and parameters well capture the real
decision process. For instance, the sets 𝐽𝑖0 , 𝐽𝑖1 , and 𝐽𝑖2 can be
determined by a survey of the TAs. The survey can have one 4. Extensions to the Basic Integer
question for each tutorial, in which each TA must choose Programming Model
one from the following three answers: (a) I like to teach this
tutorial very much; (b) This tutorial is not my favorite, but I The above basic integer linear programming model captures
can teach it if required; (c) I cannot teach this tutorial. Sets 𝐽𝑖0 , the most essential features of the TA assignment problem.
However, there are many other realistic factors that must
𝐽𝑖1 , and 𝐽𝑖2 are mutually exclusive and collectively exhaustive;
be addressed before the above model can be put into use
𝐽 = 𝐽𝑖0 ∪ 𝐽𝑖1 ∪ 𝐽𝑖2 . The teaching load allocation problem for TAs directly or put into use after minimal manual adjustment.
can be formulated as an integer linear programming model: We elaborate these factors and present approaches on how to
model these factors.
[M0]: max ∑ ∑ 𝑥𝑖𝑗
(1)
𝑖∈𝐼 𝑗∈𝐽𝑖2
4.1. Time Conflict of Two Tutorials. Usually the schedules
subject to: ∑ 𝑥𝑖𝑗 = 1, 𝑖∈𝐼 of the tutorials are a priori determined. For instance, if the
(2) course “Basics of Calculus” has 3 tutorials, then the 3 tutorials
𝑗∈𝐽𝑖
are usually scattered uniformly in a week. As a result, it is
𝑡𝑗min ≤ ∑ 𝑡𝑖 𝑥𝑖𝑗 ≤ 𝑡𝑗max , 𝑗∈𝐽 likely that one tutorial (say, tutorial 𝑖 is from 4:00 pm to 5:00
(3)
𝑖∈𝐼𝑗 pm on Monday) has time overlap with another tutorial (say,
tutorial 𝑘 is from 3:00 pm to 5:00 pm on Monday). In this
𝑛𝑗min ≤ ∑ 𝑥𝑖𝑗 ≤ 𝑛𝑗max , 𝑗∈𝐽 case, the two tutorials cannot be taught by the same TA.
(4)
𝑖∈𝐼𝑗 To address this difficulty, we define a new binary param-
eter 𝜃𝑖𝑘 , which equals 1 if and only if tutorial 𝑖 and tutorial 𝑘
𝑥𝑖𝑗 ∈ {0, 1} , 𝑖 ∈ 𝐼, 𝑗 ∈ 𝐽𝑖 . (5) have no time conflict, meaning that they could be taught by
the same TA. We add the following constraints to the basic
The objective function (1) maximizes the total number of model to formulate the time constraints:
tutorials that are taught by the most suitable TAs. Equation
(2) imposes that all tutorials are taught. Equations (3) and (4) [M1]: 𝑥𝑖𝑗 + 𝑥𝑘𝑗 ≤ 1 + 𝜃𝑖𝑘 ,
take into account the available time of each TA. Finally, (5) (6)
defines the domains of the decision variables. 𝑗 ∈ 𝐽𝑖 ∩ 𝐽𝑘 , 𝑖 ∈ 𝐼, 𝑘 ∈ 𝐼, 𝑖 < 𝑘.

Proposition 1. The above integer linear programming model Equation (6) means that, for each TA 𝑗 ∈ 𝐽, if two tutorials
cannot be solved as a linear program. 𝑖 and 𝑘 have time conflict, that is, 𝜃𝑖𝑘 = 0, then the TA can
teach at most one of them because 𝑥𝑖𝑗 + 𝑥𝑘𝑗 ≤ 1 + 𝜃𝑖𝑘 = 1.
Proof. We construct an example to show that relaxing the
integrality constraints may not lead to a correct solution. Proposition 2. In constraints (6) if there are a set of tutorials
Suppose that there are two TAs and one tutorial; the tutorial denoted by Ω such that 𝜃𝑖𝑘 = 0, ∀𝑖 ∈ Ω, 𝑘 ∈ Ω, and 𝑖 < 𝑘, then
requires two hours; the first TA is the most suitable but has we can strengthen the constraints by combining some of them
only one available hour; the second TA, with two available to one constraint:
hours, can teach the tutorial but is not the most suitable. It
is evident that the only feasible solution to the problem is to ∑ 𝑥𝑖𝑗 ≤ 1, 𝑗 ∈ 𝐽. (7)
let the second TA teach the tutorial. However, solving a linear 𝑖∈Ω
program will require the first TA to teach for one hour and
the second TA to teach for one hour. Proof. The results hold trivially based on observation.

The size of the above integer linear programming model 4.2. Repetition of Tutorials. Some tutorials may belong to
is usually not large for several reasons. First, many courses the same course, for example, “Basics of Calculus,” and in
do not have tutorials. Second, even in one department, there these tutorials TAs deliver the same contents to students. The
are several areas of specialization and TAs in one area of tutorials are repeated for two reasons. First, the tutorial class
specialization can usually only teach tutorials for courses can have a small size of students to facilitate interactions
in this area. For example, in a mathematics department, among the students and between students and the TA.
there are at least three areas: pure mathematics, applied Second, repeating the tutorials could provide students with
mathematics, and statistics. Even in the area of applied the flexibility of choosing the time slot, choosing the TA, and
mathematics, there is, for instance, financial mathematics, possibly attending the tutorial twice to improve the learning
chemical mathematics, medical mathematics, and operations outcomes. It is convenient for a TA to teach several tutorials
 5192, 2017, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2017/9057947 by Readcube (Labtiva Inc.), Wiley Online Library on [18/06/2024]. 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
Scientific Programming 5

that belong to the same course so that she could save time for Proposition 3. In constraints (13) the value of 𝑀 could be set
preparation. to |𝐼ℎ |.
To incorporate the advantage of assigning several tutori-
als that belong to the same course to one TA, we first define a Proof. The upper bound of 𝑢ℎ𝑗 is |𝐼ℎ |. Therefore, it is sufficient
weight 𝛼 to represent the benefit of assigning one more same to set 𝑀 to |𝐼ℎ | in constraints (13).
tutorial to a TA. The weight 𝛼 should be understood to be the
value of the saved preparation time by TA. We further define
a set 𝐻 to be the set of courses with at least two tutorials. The 4.3. Controlling the Number of TAs for One Course. Some
set of tutorials that belong to course ℎ ∈ 𝐻 is defined to be lecturers may want to control the number of TAs for his
course. For instance, no lecturer wants to have 12 TAs to teach
𝐼ℎ . We define new decision variables 𝑦ℎ𝑗 to be the number of
his tutorials. To reflect this requirement, we define parameter
tutorials for course ℎ ∈ 𝐻 that are assigned to TA 𝑗 ∈ 𝐽. The
𝑏ℎ as the maximum number of TAs that are assigned to the
objective function (1) should be revised to
tutorials for course ℎ ∈ 𝐻. We further define intermediate
[M2]: max ∑ ∑ 𝑥𝑖𝑗 + 𝛼 ∑ ∑ max (𝑦ℎ𝑗 − 1,0) binary decision variables 𝜋ℎ𝑗 which equal 1 if and only if TA
(8) 𝑗 ∈ 𝐽 teaches at least one tutorial for course ℎ ∈ 𝐻. We then
𝑖∈𝐼 𝑗∈𝐽𝑖2 ℎ∈𝐻 𝑗∈𝐽
could add the following constraints:
and the following constraints should be added:
[M3]: 𝑀𝜋ℎ𝑗 ≥ 𝑦ℎ𝑗 , ℎ ∈ 𝐻, 𝑗 ∈ 𝐽 (15)
𝑦ℎ𝑗 = ∑ 𝑥𝑖𝑗 , ℎ ∈ 𝐻, 𝑗 ∈ 𝐽.
(9)
𝑖∈𝐼ℎ ∑𝜋ℎ𝑗 ≤ 𝑏ℎ , ℎ ∈ 𝐻 (16)
𝑗∈𝐽
Equation (9) counts 𝑦ℎ𝑗 and the new objective function (8)
has the extra term 𝛼 ∑ℎ∈𝐻 ∑𝑗∈𝐽 max(𝑦ℎ𝑗 − 1,0). 𝜋ℎ𝑗 ∈ {0, 1} , ℎ ∈ 𝐻, 𝑗 ∈ 𝐽, (17)
It should be noted that the extra term
𝛼 ∑ℎ∈𝐻 ∑𝑗∈𝐽 max(𝑦ℎ𝑗 − 1,0) has the “max” operator that where in constraints (15) the value of 𝑀 could be set to |𝐼ℎ |
makes the model nonlinear. We propose the following because the upper bound of 𝑦ℎ𝑗 is |𝐼ℎ |.
method to linearize the objective function (8). To make the Note that constraints (16) could be changed to the
model clear, we explicitly define the new sets, parameters, following ones:
and decision variables.
󵄨 󵄨
∑𝜋ℎ𝑗 ≤ 𝑏ℎ , ℎ ∈ 𝐻, 𝑏ℎ ≤ 󵄨󵄨󵄨󵄨𝐼ℎ 󵄨󵄨󵄨󵄨 − 1 (18)
Newly Defined Sets 𝑗∈𝐽
𝐻: set of courses with at least two tutorials
𝐼ℎ : set of tutorials that belong to course ℎ ∈ 𝐻 because if 𝑏ℎ = |𝐼ℎ |, then constraints (18) are always satisfied.

Newly Defined Parameters 4.4. Controlling the Number of Days a TA Works. A TA may
like to teach five tutorials, but she may not like to teach
𝛼: benefit of assigning one more same tutorial to a TA one tutorial every day. To reflect this requirement, we define
parameter 𝜅𝑗 as the maximum number of days TA 𝑗 ∈ 𝐽
Newly Defined Decision Variables would like to teach. Evidently, 𝜅𝑗 = 1, 2, 3, 4, 5. We further
define set Θ𝑚 to be the set of tutorials taught on day 𝑚 =
𝑦ℎ𝑗 : number of tutorials for course ℎ ∈ 𝐻 that are 1, 2, 3, 4, 5, where 1 means Monday, 2 means Tuesday, and
assigned to TA 𝑗 ∈ 𝐽 so on. We define intermediate binary decision variable 𝜆 𝑚𝑗
𝑧ℎ𝑗 : an intermediate binary variable that equals 1 if and only if TA 𝑗 ∈ 𝐽 needs to teach on day
𝑢ℎ𝑗 : an intermediate continuous variable for lineariza- 𝑚 = 1, 2, 3, 4, 5. The following constraints could capture the
tion requirement regarding the number of days a TA works:
The new model is [M4]: 𝜆 𝑚𝑗 ≥ 𝑥𝑖𝑗 , 𝑗 ∈ 𝐽, 𝑚 = 1, 2, 3, 4, 5, 𝑖 ∈ Θ𝑚
[M2󸀠 ] : max ∑ ∑ 𝑥𝑖𝑗 + 𝛼 ∑ ∑𝑢ℎ𝑗
𝑖∈𝐼 𝑗∈𝐽𝑖2 ℎ∈𝐻 𝑗∈𝐽
(10) 𝜆 𝑚𝑗 ∈ {0, 1} , 𝑚 = 1, 2, 3, 4, 5, 𝑗 ∈ 𝐽
(19)
5
subject to: 𝑧ℎ𝑗 ≤ 𝑦ℎ𝑗 , ℎ ∈ 𝐻, 𝑗 ∈ 𝐽 (11) ∑ 𝜆 𝑚𝑗 ≤ 𝜅𝑗 , 𝑗 ∈ 𝐽.
𝑚=1
𝑢ℎ𝑗 ≤ 𝑦ℎ𝑗 − 𝑧ℎ𝑗 , ℎ ∈ 𝐻, 𝑗 ∈ 𝐽 (12)

𝑢ℎ𝑗 ≤ 𝑀𝑧ℎ𝑗 , ℎ ∈ 𝐻, 𝑗 ∈ 𝐽 (13) 5. Numerical Experiments

𝑧ℎ𝑗 ∈ {0, 1} , ℎ ∈ 𝐻, 𝑗 ∈ 𝐽 (14) We carry out extensive numerical experiments to demon-
strate the computational efficiency. The experiments are
and constraints (2) to (5) and (9). Note that 𝑀 in constraints implemented on a PC with 3.30 GHz of Intel Core i5 CPU and
(13) is a large positive number. 4 GB of RAM. The algorithm is coded in Matlab 2011b, calling
 5192, 2017, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2017/9057947 by Readcube (Labtiva Inc.), Wiley Online Library on [18/06/2024]. 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
6 Scientific Programming

Table 1: Computation time (s) of the models.

Number of tutorials Number of TAs [M0] [M1] [M2] [M3] [M4]

10 0.0127 0.0119 0.0139 0.0130 0.0140
20 15 0.0139 0.0146 0.0291 0.0158 0.0155
20 0.0156 0.0148 0.0300 0.0169 0.0176
15 0.0163 0.0152 0.0276 0.0189 0.0186
25 20 0.0167 0.0164 0.03675 0.0195 0.0200
25 0.0188 0.0181 0.0433 0.0221 0.0225
20 0.0192 0.0190 0.0448 0.0233 0.0239
30 25 0.0194 0.0196 0.0558 0.0245 0.0242
30 0.0210 0.0215 0.0669 0.0272 0.0281

CPLEX12.3 to solve mixed-integer linear programming mod- are formulated as mixed-integer linear programming models,
els. are further addressed to consider more practical factors.
Different combinations of the number of tutorials and This includes the following ones: (1) we consider whether
number of TAs (𝐼, 𝐽) are considered: (20, 10), (20, 15), two tutorials are delivered at the same time, and if so, they
(20, 20), (25, 15), (25, 20), (25, 25), (30, 20), (30, 25), and cannot be taught by the same TA; (2) we increase the number
(30, 30), as shown in Table 1. All of the five models [M0], [M1], of repetitions of tutorials so that it is easier for the TAs
[M2], [M3], and [M4] are evaluated for each combination. to save time for preparation; (3) a lecturer can control the
Ten random instances are generated for each model in each number of TAs for her/his class. Numerical experiments
combination, and therefore we have a total of 10 × 5 × 9 = 450 show that these models can efficiently be solved by off-the-
instances. The instances are generated as follows. In [M0], for shelf solvers, demonstrating the practical relevance of the
each tutorial-TA combination, there is 1/3 chance that the TA proposed models.
cannot deliver the tutorial, 1/3 chance that the TA can but
is not the most suitable, and 1/3 chance that the TA is the Competing Interests
most suitable for delivering tutorial; the minimum number
of tutorials that must be delivered by a TA is 0; the maximum The authors declare that there is no conflict of interests
number of tutorials that can be delivered by a TA is an integer regarding the publication of this manuscript.
uniformly drawn between 1 and 3; the minimum number of
available hours per week for a TA is 0; the maximum number
Acknowledgments
of available hours per week for TA is an integer uniformly
drawn between 1 and 5; the number of contact hours required This study is supported by the Projects of International Coop-
for a tutorial is an integer uniformly drawn between 1 and 2. In eration and Exchange of the National Natural Science Foun-
[M1], five pairs of tutorials are randomly chosen to have time dation of China (no. 5151101143) and Youth Project of National
conflict. In [M2], the number of courses is equal to half the Natural Science Foundation of China (no. 71501038).
number of tutorials and each tutorial is randomly assigned to
a course; the weight 𝛼 is set at 0.33. In [M3], the maximum
number of TAs for a course is an integer uniformly drawn
References
between 1 and 3. In [M4], the maximum number of days a [1] R. H. McClure and C. E. Wells, “Modeling multiple criteria
TA works is an integer uniformly drawn between 1 and 5. We in the faculty assignment problem,” Socio-Economic Planning
report the average CPU time required to solve one instance Sciences, vol. 21, no. 6, pp. 389–394, 1987.
for each model over the ten random instances in Table 1. It can [2] F. Y. Partovi and B. Arinze, “A knowledge based approach to the
be seen that all of the models can efficiently be solved. This faculty-course assignment problem,” Socio-Economic Planning
demonstrates the practical relevance of the proposed models. Sciences, vol. 29, no. 3, pp. 245–256, 1995.
[3] G. Hu and J. Lei, “English-medium instruction in Chinese
6. Conclusions higher education: a case study,” Higher Education, vol. 67, no.
5, pp. 551–567, 2014.
In this paper, an integer programming model is developed to [4] A. Oleson and M. T. Hora, “Teaching the way they were taught?
maximize the teaching quality by assigning most appropriate Revisiting the sources of teaching knowledge and the role of
prior experience in shaping faculty teaching practices,” Higher
teaching assistants to their tutorials. The model is very useful
Education, vol. 68, no. 1, pp. 29–45, 2014.
for teaching units as (1) it can improve teaching quality by
[5] C. Valouxis and E. Housos, “Constraint programming approach
allocating suitable teaching assistants to teach the tutorials; for school timetabling,” Computers & Operations Research, vol.
(2) the teaching allocation does not need to be manually 30, no. 10, pp. 1555–1572, 2003.
allocated, which is a time-consuming process; and (3) the [6] G. N. Beligiannis, C. Moschopoulos, and S. D. Likothanassis,
maximum teaching load of each staff will not be violated as “A genetic algorithm approach to school timetabling,” Journal of
it is modeled as hard constraints. Some extensions, which the Operational Research Society, vol. 60, no. 1, pp. 23–42, 2009.
 5192, 2017, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2017/9057947 by Readcube (Labtiva Inc.), Wiley Online Library on [18/06/2024]. 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
Scientific Programming 7

[7] N. Pillay and W. Banzhaf, “An informed genetic algorithm for

the examination timetabling problem,” Applied Soft Computing
Journal, vol. 10, no. 2, pp. 457–467, 2010.
[8] J. A. Breslaw, “A linear programming solution to the faculty
assignment problem,” Socio-Economic Planning Sciences, vol. 10,
no. 6, pp. 227–230, 1976.
[9] M. J. Schniederjans and G. C. Kim, “A goal programming model
to optimize departmental preference in course assignments,”
Computers & Operations Research, vol. 14, no. 2, pp. 87–96, 1987.
[10] M. A. Badri, “A two-stage multiobjective scheduling model
for [faculty-course-time] assignments,” European Journal of
Operational Research, vol. 94, no. 1, pp. 16–28, 1996.
[11] W. Qu, Y. Zhao, M. Wang, and B. Liu, “Research on teaching
gamification of software engineering,” in Proceedings of the
9th International Conference on Computer Science & Education
(ICCSE ’14), Vancouver, Canada, August 2014.
[12] T. F. Burgess, “Planning the academic’s workload: different
approaches to allocating work to university academics,” Higher
Education, vol. 32, no. 1, pp. 63–75, 1996.
[13] I. Vardi, “The impacts of different types of workload allocation
models on academic satisfaction and working life,” Higher
Education, vol. 57, no. 4, pp. 499–508, 2009.
[14] P. J. Bentley and S. Kyvik, “Academic work from a comparative
perspective: a survey of faculty working time across 13 coun-
tries,” Higher Education, vol. 63, no. 4, pp. 529–547, 2012.
[15] C. Halse, E. Deane, J. Hobson, and G. Jones, “The research—
teaching nexus: what do national teaching awards tell us?”
Studies in Higher Education, vol. 32, no. 6, pp. 727–746, 2007.
[16] J. C. Shin, “Teaching and research nexuses across faculty career
stage, ability and affiliated discipline in a South Korean research
university,” Studies in Higher Education, vol. 36, no. 4, pp. 485–
503, 2011.
[17] S. Hornibrook, “Policy implementation and academic workload
planning in the managerial university: understanding unin-
tended consequences,” Journal of Higher Education Policy and
Management, vol. 34, no. 1, pp. 29–38, 2012.
[18] J. D. J. Kenny and A. E. Fluck, “The effectiveness of academic
workload models in an institution: a staff perspective,” Journal
of Higher Education Policy and Management, vol. 36, no. 6, pp.
585–602, 2014.
[19] F. Li, Z. Gao, K. Li, and D. Z. W. Wang, “Train routing model
and algorithm combined with train scheduling,” Journal of
Transportation Engineering, vol. 139, no. 1, pp. 81–91, 2013.
[20] Z. Liu, S. Wang, W. Chen, and Y. Zheng, “Willingness to
board: a novel concept for modeling queuing up passengers,”
Transportation Research Part B: Methodological, vol. 90, pp. 70–
82, 2016.
[21] D. Z. W. Wang, H. Liu, and W. Y. Szeto, “A novel discrete net-
work design problem formulation and its global optimization
solution algorithm,” Transportation Research E, vol. 79, pp. 213–
230, 2015.
[22] H. Liu and D. Z. W. Wang, “Global optimization method
for network design problem with stochastic user equilibrium,”
Transportation Research B, vol. 72, pp. 20–39, 2015.
[23] B. Du and D. Z. W. Wang, “Continuum modeling of park-
and-ride services considering travel time reliability and het-
erogeneous commuters—a linear complementarity system
approach,” Transportation Research Part E: Logistics and Trans-
portation Review, vol. 71, pp. 58–81, 2014.