J. Serv. Sci. & Management, 2009, 2: 56-60
Published Online March 2009 in SciRes (www.SciRP.org/journal/jssm)
A Fuzzy Model for Evaluating Cultivation Quality of
Talents of Software Engineering at the Campus Universities
Yongzhong Lu
1
, Danping Yan
2
, Bo Liu
1
1
School of Software Engineering, Huazhong University of Science and Technology, Wuhan 430074, P. R. China;
2
School of Public
Administration, Huazhong University of Science and Technology, Wuhan 430074, P. R. China.
Received January 6
th
, 2009; revised February 9
th
, 2009; accepted March 2
nd
, 2009 .
ABSTRACT
In order to measure the quality of talent cultivation at the school of software engineering, a quality evaluation model
based on fuzzy theory is put forward. In the model, a three-layer architecture, which is composed of overall goal layer,
second goal layer, and attribute layer, is set up. It places emphasis on the demand of talents with practicability and
engineering in the field of software engineering. Then a case is used in the model to illustrate its effectiveness. The ex-
perimental results show that the model can comparatively better evaluate the quality of talent cultivation, reach the
expected objective, and fulfill the practical demand. According to the model, a quality evaluation software system is
developed while a rainfall lifecycle development model and Microsoft Visual C++ Development Studio are utilized.
Keywords: s oftware engineering, cultivation quality evaluation, fuzzy computing model
1. Introduction
In order to fulfill the urgent social demands of software
talents with high quality, practical experiences and com-
prehensive engineering skills in China, we have carried
out a series of reform and innovation pertaining to the
teaching contents and approaches, courses system, and
management institution and operational mechanism. Up
to now, we have come to deeply recognize that training
talents of software engineering is similarly deemed to a
item of talent production project. In the course of the
teaching reform and innovation, it is significantly vital to
lay emphasis on its training quality and effect which are
the progress signpost in the forthcoming days. Generally
speaking, it is rather difficult to measure the quality and
effect of bringing up software talents quantitatively be-
cause they are closely related to numerous determinants
[1,2]. Therefore, an accurate quality evaluation model
about the training project of software talents at the uni-
versities is still not set up. Based on the social demands
for software talents in China, we first put forward a
qualitative model of quality evaluation of talent cultiva-
tion at the universities, and then exploit a fuzzy approach
to give the quantitative computational results. Subse-
quently a case is used to testify its effectiveness. At last, a
quality evaluation software system is developed while a
rainfall lifecycle development model and Microsoft Vis-
ual C++ Development Studio are utilized.
2. A Fuzzy Quality Evaluation Model for
Software Talents
We have referred to the generic ability evaluation standard
of engineering graduates in UK, the USA and other
European developed countries [3,4,5,6,7,8]. In addition,
we have combined it with present practical situation at
the campus schools and amended it properly. As a result,
a quality evaluation model of training software talents is
presented in Figure 1. In the model there are three layers:
the top one is called overall objective layer and expressed
by matrix A, the middle layer is called second objective
layer and expressed by matrix B, and the lowest layer is
called third attribute layer and expressed by matrix C, but
it does not mean this layer is no importance. The corre-
sponding statements are shown in Table 1.
3. A Fuzzy Evaluation Approach
It’s quite difficult to get the exact values of the attributes
in the model above. The fuzzy evaluation approach
adapts to solve the problem well. Therefore it is used here
to work out the solution to the problem. Its process is
described as follows.
1) Establish the evaluation expert group
Different types of software experts are adopted to
probe into the quality of training the software talents.
They are usually composed of several experts such as
field experts, senior managers, and users, and so forth.
After the selection of evaluation expert group, a comment
set is required to be determined. Supposing that the hier-
archical rank of software products is classified into five
levels which correspond to a comment set V: V= (“excel-
lent”, ”good”, ”medium”, ”passed”, ”bad”) =(v
1
,v
2
,v
3
,v
4
,
v
5
).

YONGZHONG LU, DANPING YAN, BO LIU 57
Figure 1. An evaluation model of training the talents of software engineering
Table 1. The generic statements corresponding to the Figure 1
1 Ability to exercise Key Skills in the completion of software engineering-related tasks at a level implied by the benchmarks
associated with the following statements (B
1
)
a) Communication (C
11
)
b) Information Technology (C
12
)
c) Application of Number (C
13
)
d) Working with Others (C
14
)
e) Problem Solving (C
15
)
f) Improving Own Learning and Performance (C
16
)
2 Ability to transform existing software systems into conceptual models (B
2
)
a) Elicit and clarify client's true needs (C
21
)
b) Identify, classify and describe software engineering systems (C
22
)
c) Define real target software systems in terms of objective functions, performance specifications and other constraints (ie, de-
fine the problem) (C
23
)
d) Take account of risk assessment, and social and environmental impacts, in the setting of constraints (including legal, and
health and safety issues) (C
24
)
e) Resolve difficulties created by imperfect and incomplete information (C
25
)
f) Derive conceptual models of real target software systems, identifying the key parameters (C
26
)
3 Ability to transform conceptual models into determinable models (B
3
)
a) Construct determinable models over a range of complexity to suit a range of conceptual models (C
31
)
b) Use mathematics and computing skills to create determinable models by deriving appropriate constitutive equations and
specifying appropriate boundary conditions (C
32
)
c) Use industry standard software tools and platforms to set up determinable models (C
33
)
d) Recognise the value of Determinable Models of different complexity and the limitations of their application (C
34
)
4 Ability to use determinable models to obtain system specifications in terms of parametric values (B
4
)
a) Use mathematics and computing skills to manipulate and solve determinable models; and use data sheets in an appropriate
way to supplement solutions (C
41
)
b) Use industry standard software platforms and tools to solve determinable models (C
42
)
c) Carry out a parametric sensitivity analysis (C
43
)
d) Critically assess results and, if inadequate or invalid, improve knowledge database by further reference to existing software
systems, and/or improve performance of determinable models (C
44
)
5 Ability to select optimum specifications and create physical models (B
5
)
a) Use objective functions and constraints to identify optimum specifications (C
51
)
b) Plan physical modelling studies, based on determinable modelling, in order to produce critical information (C
52
)
c) Test and collate results, feeding these back into determinable models (C
53
)
6 Ability to apply the results from physical models to create real target software systems (B
6
)
a) Write sufficiently detailed specifications of real target software systems, including risk assessments and impact statements
(C
61
)
b) Select production methods and write method statements (C
62
)
c) Implement production and deliver products fit for purpose, in a timely and efficient manner (C
63
)
d) Operate within relevant legislative frameworks (C
64
)
7 Ability to critically review real target software systems and personal performance (B
7
)
a) Test and evaluate real software systems in service against specification and client needs (C
71
)
b) Recognise and make critical judgements about related environmental, social, ethical and professional issues (C
72
)
c) Identify professional, technical and personal development needs and undertake appropriate training and independent re-
search(C
73
)
C
21
-C
26
C
31
-C
36
C
41
-C
46
C
51
-C
56
C
11
-C
16
C
71
-C
76
C
61
-C
66
B
2
B
3
B
4
B
5
B
1
B
7
B
6
Overall objective layer
second objective layer
third attribute layer
Comprehensive evaluation of training the talents of software engineering ( A)

58 YONGZHONG LU, DANPING YAN, BO LIU
2) Determine the single weights of the statements
AHP (Analytical Hierarchy Process) is adopted to fig-
ure out the weights of the statements. The detailed steps
are followed below.
 According to the model above, a proper questionnaire
is well-prepared for the experts. They determine the
mutual weights among the statements in three layers.
The weight matrix between overall objective layer A and
second objective layer B
i
is shown in Table 2. The matrix
is usually called determinant matrix. We can obtain other
determinant matrixes in the same way. Thereafter they
fill out the comments about the attribute layer statements
as Table 3.
 Construct the single determinant matrix
The AHP constructs the determinant matrix by terms
of relationship among the statement items, and their pro-
portional scales are among 1-9 [9]. Supposing that A
represents the object set, U the evaluation item set, u
i
(i=1,2,…,n) the evaluation item, and u
ij
represents mutual
weight between u
i
and u
j
(j=1,2,…,n), the determinant
matrix is expressed below.













⋅
nnnjnn
nj
nj
n
ni
uuuu
uuuu
uuuu
u
u
u
uuuuU
21
222221
111211
2
1
21
LL
LL
M
LL
(1)
 Calculate the normalized weights of all evaluation
items above
The geometric average method is used to gain the ei-
genvector corresponding to the most characteristic root
max
λ of matrix U above. And it is normalized and shaped
into the weights of all evaluation items. The detailed
formula is following
∑∏∏ ===
=
n
i
n
n
jij
n
n
jiji
uuW
1
1
1
1
1
)()(
(2)
where i, j = 1, 2 ,…, n. The result
T
n
WWWW ),,,(
21
L=
is the above-mentioned eigenvector.
Table 2. Weight matrix of A and B
A B
1
B
2
B
3
B
4
B
5
B
6
B
7
B
1
B
2
B
3
B
4
B
5
B
6
B
7
Table 3. Subjection degrees about attribute layer statements
Comment set
Attributer layer
statements excellent
good Medium passed bad
C
11
C
12
C
73
 Consistency testing
Supposing that U is a matrix with n ranks, u
ij
(1≤i≤n，
1≤j≤n) is an element in U, if all elements of U have a
property of transitivity, that is to say
ikjkij
uuu=×, the
matrix U is called a consistency matrix. A consistency
matrix can be verified by the formula (3)
RICICR = (3)
where CR is called the random consistency ratio of the
determinant matrix, RI is called the average random con-
sistency ratio of the determinant matrix, and CI is called
the general consistency item which can be expressed by
the formula (4)
)1()(
max
−−= nnCI λ (4)
where n is the rank of the determinant matrix.
max
λ is
decided by the following formulae (5) and (6)
∑
=
=
n
ii
i
W
PW
n
1
max
)( 1
λ (5)


























=












=
n nnnjnn
nj
nj
n
W
W
W
uuuu
uuuu
uuuu
PW
PW
PW
PW L
LL
LL
LL
LL
L
2
1
21
222221
111211
2
1
)(
)(
)(
(6)
when CR<0.10, it can be concluded that the determinant
matrix has a satisfactory property of consistency, that is
to say that the distributed weights are proper, vice versa.
 Calculate the comprehensive weights
The distributed weights of the second objective layer to
the third attribute layer are obtained by the formula (3).
The distributed weights of the overall objective layer to
the second objective layer is calculated by the formula (7)
ij
n
jj
WCWBW
∑
=
=
1
(7)
where
j
WB is the important weight of
j
B (1<j<7)
corresponding to A, and
ij
WC is the important weight
of
ij
C corresponding to
j
B . When
j
B has no bearing
with
ij
C ,
ij
WC =0.
3) Determine the subjection degrees of the quality
evaluation
When carrying out the evaluation of talent cultivation
of software engineering, field experts, together with sen-
ior manager (policy-makers) and customers, give the de-
cisive subjection degree according to the defined com-
ment set above. It can explicitly be expressed by the sub-
jection degree matrix R below
kmij
rR
×
= )( (8)
where
ij
r is the percentage of regarding the i-th evalua-
tion statement as the j-th comment class. And it is also

YONGZHONG LU, DANPING YAN, BO LIU 59
expressed by ddr
ijij
= where
ij
d is the number of
the members of drawing the conclusion that the i-th
evaluation statement belongs to the j-th comment class, d
is the total of the members, m is the number of the state-
ments, and k is the evaluation rank.
4) Calculate the final evaluation result
After attaining the subjection degree matrix R, we cal-
culate the comprehensive evaluation vector S of talent
cultivation of software engineering. Then we adopt the
Weighted Average Model of comprehensive evaluation-
M (*,+) in order to consider all relevant factors appropri-
ately and remain their information. The comprehensive
evaluation vector S and the comprehensive evaluation
result P are displayed in (9) and (10) respectively
RWS
a
c
×= (9)
T
SVP ×= (10)
In the formula (9),
a
c
W is the comprehensive weights
of third attribute layer C corresponding to overall objec-
tive layer A. As a result, the quality level of talent culti-
vation of software engineering at campus universities can
easily be performed by the formula (10) and the task of
quality evaluation of talent cultivation of software engi-
neering is successfully completed.
4. Illustration
In order to testify the effectiveness of the presented
model above, we take a practical case for example. Based
on the model, we perform the demonstration in accor-
dance with the following steps.
1) Calculate the single weights of the statements
The AHP is exploited to construct the single determi-
nant matrixe as Table 2 and normalized by Formula (2).
Then the consistency testing is done by Formula (3). If
the CR is less than 0.1, the comprehensive determinant
matrix is obtained by Formula (7). The two results are
shown in Tables 4 and 5.
2) Calculate the subjection degree matrix
After the mutual weights of three layers are decided,
15 relevant members give their evaluation opinions to the
quality of talent cultivation of software engineering with
the aid of the comment set above. The subjection degree
matrix
530 ×
R is gotten by Formula (8) and normalized
into Formula (11).
















=
L
L
L
L
L
0000000
0091.00091.0000
564.0273.0273.0345.0091.00564.0
455.0345.0273.0273.0345.0564.0455.0
182.0091.0455.0081.0564.0636.0195.0
R
(11)
3) Calculate the comprehensive evaluation value
)067.0,0415.0,2123.0,4635.0,2175.0(=
×= RWS
a
c
(12)
728.3
)067.0,0415.0,2123.0,4635.0,2175.0()1,2,3,4,5(
=
×=
×=
T
T
SVP
(13)
From Formulae (12) and (13), we find that if the subjec-
tion degree is 0.2175, the quality is excellent; if the
Table 4. The single weights of the second objective layer
B corresponding to the third attribute layer C
B
1
(0.476) B
2
(0.266) …
C
11
C
12
C
13
C
14
C
15
C
16
… …
0.299 0.141 0.105 0.168 0.127 0.160 … …
Table 5. The comprehensive weights of the second objec-
tive layer B and the third attribute layer C correspond-
ing to the overall objective layer A
B
1
(0.476) B
2
(0.266) …
C
11
C
12
C
13
C
14
C
15
C
16
… …
0.138 0.089 0.049 0.078 0.059 0.096 … …
Figure 2. The system workflow
Open file
Add or Delete data
Acquirement of determi-
nant matrix of attribute
layer
Acquirement of determinant
matrixes of overall objective
layer and second objective layer
and judgement of consistency
Calculate the single evalua-
tion value
Calculate the characteristic
vectors of overall objective layer
and se
c
ond objective
layer
Calculate the final evaluation result
Save file

60 YONGZHONG LU, DANPING YAN, BO LIU
Figure 3. The quality evaluation system
subjection degree is 0.4635, the quality is good; if the
subjection degree is 0.2123, the quality is medium; if the
subjection degree is 0.0415, the quality is passed; if the
subjection degree is 0.067, the quality is bad. If
=
V
{
}
1,2,3,4,5 is quantified, the comprehensive evaluation
value is 3.728 and its final evaluation quality is “medium”.
5. Developing the Quality Evaluation System
The workflow of the quality evaluation system is de-
scribed as Figure 2. In the figure, we divide the system
into five modules which include Add or Delete module,
Calculate the single evaluation value of certain attribute
module, Consistency testing module, Calculate the char-
acteristic vector module, and Calculate the final evalua-
tion value module.
The quality evaluation software system is developed as
Figure 3 while a rainfall lifecycle development model and
Microsoft Visual C++ Development Studio are utilized.
6. Conclusions
Based on the quality evaluation model of talent training
of software engineering, a fuzzy quality evaluation sys-
tem of talent training of software engineering is devel-
oped. It can easily measure the quality level of talent cul-
tivation of software engineering and provide a good
evaluation platform for software talent cultivation. How-
ever, some aspects on the consistency testing and deter-
minant matrix construction will be further addressed in
the future.
7. Acknowledgement
The support from the Natural Science Foundation at
Huazhong University of Science and Technology grant
number 2007Q006B is gratefully acknowledged for this
work by the authors.
REFERENCES
[1] J. S. J. Lin, “Study of university curriculum development
for human resource management and the expectations of
business managers,” International Journal of Business and
Systems Research, 2(4), pp. 418-430, 2008.
[2] P. Xiao, Y. Cheng, and Y. Yang, “Higher education hu-
man resources development criterion inquisition,” Future
and Development, 29(2), pp. 9-12, 2008.
[3] N. E. Gibis, “The SEI education program the challenge of
teaching future software engineers,” Communications of
ACM, 32(5), pp. 594-605, 1989.
[4] O. Hazzan, “The reflective practitioner perspective in
software engineering education,” Journal of Systems and
Software, 63(3), pp. 161-171, 2002.
[5] A. J. Cowling, “What should graduating software engi-
neers be able to do?” IEEE Proceedings of the 16th Con-
ference on Software Engineering Education and Training,
pp. 88-98, 2003.
[6] D. T. Holt and Mitchell Crocker, “Prior negative experi-
ences: their impact on computer training outcomes,”
Computers & Education, 35(4), pp. 295-308, 2000.
[7] M. M. Boulet, C. Dupuis, and N. Belkhiter, “Selecting con-
tinuous training program and activities for computer profes-
sionals,” Computers & Education, 36(1), pp. 83-94, 2001.
[8] D. Garlan, “Making formal methods education effective
for professional software engineers,” Information and
Software Technology, 37(5-6), pp. 261-268, 1995.
[9] Q. Wang, “Practical fuzzy mathematics,” Science and
Technology Documentary Press, Peking, 1995.
(Edited by Vivian and Ann)