Accurate Knowledge Evaluation by Deep Datamining in Telecommunication Engineering Studies

Roberto Llorente#1, Maria Morant#2

#

Universidad Politcnica de Valencia, C/ Camino de Vera s/n, 46022 Valencia (SPAIN)

1
rllorent@dcom.upv.es,
2
mmorant@upvnet.upv.es

system-analysis tools such as convolutions, the impulse

response and domain transformations like Fourier and Laplace
transforms applied to system analysis.
The ASC subject comprises 75 lecturing hours and
provides the theoretical basis (80 % time) and laboratory
lessons (20 % time). The laboratory sessions consist of a
discovery process in which if-then cases are evaluated. The
aim is to try different mathematical operations (using
MATLAB) to extract their own conclusions and to realize
about the implications of the concepts presented in the
theoretical lessons. In the laboratory, the alumni also learn
how to use the instrumentation and the associated techniques
to be employed in the system analysis, focused on electronics
circuits. The practical work is thoroughly guided in the first
parts but, eventually, the student must become autonomous in
the measurements. The laboratory time is intended to practice
I. INTRODUCTION
theoretical concepts previously introduced in the theoretical
Nowadays there is a growing interest in the use of on-line
lessons. The student final qualification includes the practice
tools, Internet-based or in a local area network, for interactive
marks obtained in the laboratory sessions.
learning. There is also a rising interest in cooperative activities
Using an ordinary exam in the laboratory sessions to track
leading to support the students and, in particular, cooperative
the progress is time-consuming and produces lack of interest
learning. These two lecturing techniques are inline with the
in alumni due to the large number of sessions along the term.
new challenges for the European Higher Education area to be
Also the number of students per session (around 35) prevents
created in 2010 based on the so-called Bologna spirit [1].
oral or individual exams. The proposed evaluation technique
This paper proposes the novel approach of combining these
for laboratory lessons consists of an on-line exam to be done
two elements, on-line learning tools and groupal cooperative
at the end of each laboratory session. This on-line application
learning, to foster the knowledge acquisition by the alumni.
was designed and implemented by lecturing staff of ASC, and
These two elements, combined, permit an effective learning
introduced during the academic year 2005/2006 with excellent
process enabled by the precise and accurate tracking of the
results. Further work on this application includes a control
learning process. This tracking is performed by the collection
panel system that advices the lecturer in almost real-time with
of several variables during the on-line learning process, and
the student success rate statistics broke down by the concepts
also in the groupal cooperative work. All this information is
introduced in the session. For this purpose, the results of the
processed by deep data-mining, i.e. a large number or
on-line exam are stored in a data-base and processed on-thevariables are considered, with two key outcomes: First, an
fly through data-mining.
accurate estimation of the effectiveness of the learning process.
On the other hand, interactive seminars are included as a
And second, a precise identification of the bottlenecks of areas
part of the theoretical lessons to apply the explained concepts
where the student is facing problems in the knowledge
to solve a proposed problem. Using this method, students
acquisition process. This is of great importance in order to
resolve proposed problems first as a team in the seminars,
facilitate the lecturer to take a corrective action.
then with a partner in the laboratory sessions, and finally
The proposed techniques have been implemented in the
during the lecturing via examples and in-class resolved
subject Anlisis de Sistemas Contnuos (ASC), belonging to
problems. This makes easier to the student to prepare this very
the curricula of Telecommunication Engineering Degree
problem-solving oriented subject.
offered by the Escuela Politcnica Superior de Gandia (from
This paper is divided in four sections: the first Section
the Universidad Politcnica de Valencia). ASC is a one-term
describes the proposed evaluation technique for laboratory
subject considered a keystone in Telecommunication studies,
lessons in engineering studies. The second Section depicts the
as it describes basic time-domain operations concerning
concepts for the interactive groupal seminars for theoretical
Abstract This paper describes a novel approach for the
evaluation of the alumni knowledge acquisition based on deep
data-mining techniques. Cooperative learning and data-mining
can be combined to foster the knowledge acquisition process of
the students. This approach has been introduced in the subject
Anlisis de Sistemas Contnuos in the second term of the first
year of the Telecommunication Engineering Degree studies at the
Escuela Politcnica Superior de Gandia from the Universidad
Politcnica de Valencia (UPV), Spain. The evaluation method is
based on laboratory lessons and seminars. The laboratory lessons
include an on-line exam with a specific set of variables targeted
for data-mining processing. Interactive groupal seminars are also
introduced in the subject. In this case, the accurate knowledge of
the skills acquired by individual students is evaluated by which
decorrelation techniques on a large set of variables which are
also processed in the data-mining.

978-1-4244-5386-3/09/$26.00 2009 IEEE

basis. The obtained results of using these techniques are

shown and further work is described in Section IV. Finally in
Section V the main conclusions are drawn.
II. LABORATORY ON-LINE EVALUATION WITH DATA-MINING
An e-learning tool has been developed targeting to track the
process and evaluate the work done in the laboratory. The
main target is to solve the low assistance and relatively poor
results obtained in laboratory sessions dealing with Fourier
and Laplace domains in telecommunications engineering. The
developed tool consists of a web-based class-level on-line
exam, implemented through an examination server which
includes data-mining functionalities.
Basically, data-mining is concerned with the analysis of
data and the use of software techniques for finding patterns
and regularities in sets of data. A lot of companies use datamining to comb through databases to figure out which of their
products sells best and where. In Fig. 1 it can be observed the
on-line proposed technique and the relationships between the
information used for data-mining.
Record
On-line
test server

Data mining
DATA
Control Panel

On-line test

Fig. 1. Proposed data-mining technique in an on-line evaluation.

The special importance of the proposed technique is that

data-mining is done at class-level and also at student-level.
This permits the proper guide of the knowledge building
process at personal level. The implemented tools combine
adaptive examination with data-mining techniques in the
e-learning examination phase. The on-line exam and control
panel application for data-mining are described now:
A. Web-based Exam and Data Collection strategy
The students are presented with an exam whose workflow
is summarized in Fig. 2. The examination consists of an online exam for groups of two or three students to be done in the
last 15 minutes of every laboratory session.
Identification
Start Exam

Question
Database

Question
Previous
Next

No

Finish?
End of time

Yes

Results
Database

Final Results

Fig. 2. Workflow of the on-line exam presented to the students.

The exam is composed on-the-fly by an examination server

that chooses the questions from a question database, and saves
the student results in a result database (answers, spent time
and other variables are recorded). The results database is
continually monitored in order to track the success rate of the
different concepts addressed in the exam.
The examination approach can be summarized in:
Alumni are presented to an on-line exam consisting of
five questions covering the subjects studied in the
laboratory session. The questions presented are selected
by the exam server following these guidelines:
a. The exam is composed on-the-fly by the
examination server from a questions database
(XML). Each entry in the question database
consists of text, answer, difficulty, recurrence,
average student time spent and success rate [2].
Every question stored in the question database is
tagged indicating the concept addressed and its
estimated difficulty (from 0 to 10). One question is
selected for each main concept studied in the lab
session, i.e. only five main concepts can be
addressed by session. This is coherent with the
session duration, 2.5 hours. The questions are
selected randomly from the XML question database,
at this initial stage.
b. The questions presented are sorted in increased
difficulty order. As the order is very step, some
questions can be re-selected from the XML
database.
c. A recurrence index is stored in order to guarantee a
fair distribution of the questions presented.
Every question must be solved (on-line) after
calculations, measurements or programming in
Matlab, if required. After each question, the selection
is automatically recorded in the results database. The
recorded data includes the answer, if it is correct/wrong,
and other information for data-mining.
The questions presented to each group of students are
completely different. In this way, the students are free
to move around the laboratory consulting the reference
materials, using the simulation tools and/or the
measurement equipment. They can even cooperate to
solve their different questions.
The students count with the help of a class-tutor, who
guided the laboratory work, which can provide answers
to technical questions even during the on-line
evaluation.
This e-learning tool for laboratory evaluation gives several
advantages to the teaching/learning process:
1. The students keep asking and learning along all the
laboratory session, including the exam time.
2. They start to cooperate to solve the questions
(cooperative working) and use materials and equipment in
order to get a conclusion by themselves. This is of great
importance as it is the basis of research.
3. The students keep also more interested during the
laboratory session. The exam difficulty gradation facilitates to

reach the maximum qualification possible. They receive the

results and correct answers just when the exam finishes, which
facilitates learning due the instant feedback.
Moreover, the alumni performance is identified by realtime data-mining on the students answers. No all the students
do the exam at the same time, it is not necessary as the
questions presented to each student are different. The
data-mining from an actual examination session takes into
account data from previous examination questions, so
concepts with minima results concepts are already identified.
As the server records the time spent and success rate of
each question, the average difficulty of each question is
analyzed and difficulty information is stored in the question
database.
B. The Control Panel
The procedure for correcting the assessment on-line tests
presented is, naturally, the individuation of right and wrong
answers. The control panel (CP) system applies tailored deep
data-mining [3] to track the knowledge construction process
of the alumni. This CP is a translation of the business control
panel that can be found in advanced business administration
techniques [4]. This aims to monitor the results and the quality
of the learning in the laboratory.
Usually the computer is the responsible for finding the
patterns by identifying the underlying rules and features in the
data. In this case, data-mining is applied over a large set of
variables which are tracked during a simple on-line exercise
done at the end of the laboratory lesson.
Once the alumni have entered its login, the server
authenticates if they are allowed checking the list of
matriculated alumni. To start the exam, the test server records
in the results database a new entry of exam initialization with
the following information:
Name of the user that has opened the UPV connection
in the computer with its UPV login.
Data and hour of the event.
IP address of the computer.
Name of the XML archive is running. This is the name
of the current exam file containing the different
questions saved in the database of the examination
server. It will be at least as XML files as lab sessions.
Name of the alumni. The database has 3 columns
reserved for the student names.
Event and data. There are two event types defined. The
event Exam Initialization consists of a row that in the
data column includes the selected questions of the
current test. The other event, Exam Ending, shows in
the data column the number of succeeded, failed and
blank answers.
Marks. Just in case the event is Exam Ending the
server records the calculated mark for the test.
With this information several statistics for the data-mining
process are obtained. These are summarized in:
Ts: Time spent by the student observing the question
on-screen. This value can give the professor an idea
concernig if the group has discussed about the answer
selection a lot of time. This parameter can have several

meanings. For example, if the spent time is short and

the answer is correct the theoretical concept that the
question is addressing is well understood. If not, this
concept should be recalled in the next lesson to be
clarified. Another meaning could be if the question is in
a wrong format. This monitoring can give the professor
the opportunity of giving the students another question
in the same topic to observe the results and explain the
concept again if necessary.
Tr: Number of times which the student has recalled the
question to be presented on the screen. This parameter
includes at the background the internal relationships
between the questions presented to the students. If a
question is recalled after answering it, the reason could
be because the students have understood the concept
after solving a question that was presented afterwards,
and they want to change the previous selection.
Ai: Number of times the studies has changed the answer.
This parameter is related with the two previous ones.
The students can spend some time observing the
question and if the selected answer is changed many
times it indicates that there is no agreement of all the
members of the group. This variable is related with Tr if
the question is recalled after answering.
Seq_a: Sequence of answers selected by the student
when changing the answer of a given question. This is a
very important parameter to be considered. Data-mining
studies can point out if any question was wronganswered but if the previous selected answer was the
correct one. Moreover, it is interesting to consult this
parameter if a question has been recalled after
answering it, because most of the times when a test
answer is changed the previous was the correct one.
Seq:q: Sequence of questions requested by the student
after the first presentation. This parameter contains the
questions the students wanted to recall after answering
the complete test.
Lq: Last question answered by the student.
S1 SN: Individual scores per question.
S: Final score.
The correlation of the different variables above is analyzed
in order to evaluate the average mark and the number of
incidences and to identify the bottlenecks or roadblocks in the
student learning process. This is taken into account by ASC
lecturing staff and leads to several corrective lecturing actions
which improved the evaluation results as it will be described
in Section IV.
III. INTERACTIVE GROUPAL SEMINARS
This Section describes the interactive groupal seminars
proposed for theoretical basis in engineering studies. These
seminars are based on collaborative work to solve a proposed
problem in a one-hour session every week.
Working in group provides opportunities for developing
student generic skills such as organization, team work,
delegation and cooperation. Group work can be used in
engineering studies for introducing the students in real world

work as it provides the opportunity to work in

multidisciplinary teams. The aim of group work is to produce
better results in presentations and reports. This is achieved
combining the individual talent of each group member,
contributing knowledge and ideas.
Cooperative work is not achieved only having students sit
side-by-side at the same desk. To be cooperative, a group
must have clear positive interdependence: members must
promote each other's learning and face-to-face interaction.
Interaction between teacher and students and between students
plays a fundamental role in the whole learning process. In
many cases, also the practical application in the laboratory or
in seminar activities of the notions acquired by the students
constitutes a very important step.
Problem-based learning is the result from the process of
working toward the understanding or resolution of a problem.
This is very suitable for engineering and for medicine, where
it is currently used, because it helps students develop skills
and confidence for formulating problems. This activity results
in learning the process of building models to solve problems
together in face-to-face interpersonal interaction which is
difficult to achieve in any other way [5].
The groupal seminars are developed in four phases:
In the first part of the seminar, the alumni listen to the
explanation of theoretical concepts related with the problem
that will be proposed afterwards. This focuses student
attention on the material to be learned. In this phase, the
professor clearly defines the assignment, teaching the required
concepts and strategies. It is needed to specify the positive
interdependence and individual accountability and give the
criteria for success. In the second phase, the students work in
group trying to solve the problem. The group size is a factor to
take into account. Smaller groups (of three students) contain
less diversity of thinking styles and make more difficult the
management (as it means that there will be a larger number of
groups in the same class). Conversely, in larger groups it is
difficult to ensure that all members participate [6].
The second phase needs the organization of the class to be
changed as shown in Fig. 3. Effective classroom organization
maximizes the time spent on teaching and learning by
eliminating or minimizing the distractions caused by behavior
problems.
S

c
f

Fig. 3. Plan view of the organization of the class for the different phases of
the seminar.

Professor
explanation

Work in groups to
solve the problem

Give the problem

to the professor

Solution explanation
and problem check

Fig. 4. Phases of the proposed group seminar technique.

During the first phase of the seminar, the classroom

organization remains as in a usual lesson, in order to make
easier to the students attending the professor explanation.
After the explanation, the placement of furniture is changed
grouping up to 5 students in small clusters of desks to form
different groups of work. This provides a separated area of
work for each group of students (that facilitates small group
cooperative learning) and includes ease of movement around
room for teacher.
In the third step of the seminar, the students give the
solution of the proposed problem to the professor, that will
return it solved back in the next session.
The fourth and last phase consists on the exposition of the
problem solution by the lecturer.
This technique can be used by the lecturers to monitor the
quality of the results. In the last phase of the seminar, the
lecturer can go deeper in the concepts that were not clear
during the development of the problem inside the group. A
concept diagram illustrating the different phases of the
proposed seminar technique is shown in Fig. 4.
In most cases, a group is comprised by students with
different levels of ability, and it is a challenge for the lecturer
to evaluate individually each member. This can be solved
using de-correlation techniques in the evaluation process. The
herein proposed technique targets to obtain a deeper
knowledge of the skills acquired by de-correlating the group
evaluation scores evaluation from the individual scores by
different approaches:
A. Group score
The first decorrelation technique evaluates the results
obtained by the group during the groupal seminar. In this
score all group members are pooled together, taking into
account the quality of the intermediate steps employed to
solve the problem (Si) and if the correct solution of the
problem is achieved (Ss). Including this score in the
evaluation process aims to promote a positive interdependence
inside the group, as the team members perceive that they need
each other to complete the group's task. This avoids the effect
of pseudo-learning group. The called pseudo-learning group
effect occurs when students are assigned to work together but
they have no interest in doing so. Sometimes students think
that they will be evaluated by being ranked from the highest
performer to the lowest performer, so they start competing
between them. They see each other as a rival who must be
defeated, and in this case, students would achieve better
results if they were working alone.
However, it would be not fair using only this score to
evaluate the students, as it remains the commented possibility
when the work was left to one or a few students of the whole

group. Cooperation is not assigning a task to a group of

students where one or only a few students do all the work and
the others only put their names on the report as well. This is
the reason why de-correlation techniques are needed to
evaluate the individual members of the group.
B. Individual participation
A second de-correlation technique is based on evaluating
the participation of each student inside the group. This
promotes face-to-face interaction, assessing the quality and
quantity of each member's contributions to the group.
During the second phase of the seminar, the professor
observes each group and records the frequency with which
each member contributes to the group's work. Based on this
information, the lecturer assigns an individual score (Sp) to
each student depending on its participation inside the group.
The final score is given by the score obtained by the group
presented solution weighted by the individual score
Sf=Sp*(Si+Ss).
C. Peer-to-peer student evaluation
The third technique takes into account peer-to-peer student
evaluation at group level. Professors need to ensure that all the
members of the cooperative learning group are involved in the
task. Groups need to describe what member actions were
helpful and unhelpful.
For this technique, each student gives a score to each
student inside the same group. The professor provides time
and a punctuation structure for the students of each learning
group. This reflects how effectively they have been working
together. When a given score is recorded for the same student
in different groups, this correction factor is accurate.
D. Percentile evaluation
The last technique is based on a percentile evaluation for
the group and session. In order to avoid conflicts in the group
where some students can be too domineering or don't do their
share of the work, the student groups should be different in
each groupal seminar session. It is important to change student
groupings frequently. This ensures that each student interacts
with different students throughout the semester.
The individual scores are extracted from the percentile of
the group scores over all the sessions on different group
environments. If a large number of scores are recorded, the
final score of each student is precisely the percentile above the
mean of the class.
IV. RESULTS AND FURTHER WORK
After the development of these evaluation techniques,
several bottlenecks in the subject were identified. Corrective
lecturing actions, like allocating more time for the Fourier
Analysis lecturing, were introduced and the pass-rate of the
laboratory work in the subject increased in 37% compared
with the first year the laboratory work was introduced
(academic year 2003/2004). This improvement has been
sustained over time, as can be observed in the final marks.
The historic series for the last four academic years are shown
in Fig. 5.

2004/2005
Pass
Fail
38.8% 61.2%
2006/2007
Pass
Fail
29.4% 70.6%

2005/2006
Pass
Fail
29.8% 70.2%
2007/2008
Pass
Fail
23.4% 76.6%

Fig. 5. Pass/fail results, before and after introducing the herein proposed
evaluation techniques.

Further enhancements will be included in the laboratory

on-line evaluation application in the next years. The
parameters used for data-mining could be improved to obtain
more information about the recorded results. Moreover, other
data-mining solution could be used to make the process more
flexible and dynamic. Several solutions are being studied for
its further selection. Nowadays it seems that Clementine
software could fit with the application that is implemented.
Clementine is data mining solution that helps organizations
understand peoples past behavior and predict future behavior.
Using Clementine, we could access data from various sources,
not only from Oracle, and evaluate and deploy analytical
models. The predictive analytics is an attractive approach to
be considered.
It should be pointed out that the introduction of the on-line
evaluation tool in the laboratory in 2005/2006 made a 32%
increase in attendance to the laboratory sessions, and a 9%
increase in alumni passing the subject. In 2007/2008 with the
data mining tool a 6% increase was observed.
Table I shows an example of the obtained results for a case
study of 16 students in a seminar sessions to be compared
with the results of the laboratory lesson. This table contains
several data for each student as:
Group: Four groups were formed in each seminar
session (labeled 1, 2, 3 and 4, respectively) in order to
obtain the mean score of the group and the percentile
study. It can be observed that each student is in a
different group in each session to ensure de-correlation
in the percentile evaluation described before.
Si: Score due to the quality of the intermediate steps
used by the student to solve the problem.
Ss: Score of the problem considering if the correct
solution was achieved.
Sp: Individual participation inside the group. This value
is recorded by the professor during the seminar session.
This parameter has a maximum value of 0.5.
Sf: Score obtained by the group presented solution
weighted by the individual score Sf=Sp*(Si+Ss).
P2P: Peer-to-peer score given by each student to the
other students inside the same group. The lecturer asks
the students to record each score with an integer number.
Final: Final score of the seminar session. This score is
weighted in this way: Final=0.6*Sf+0.4*P2P.

TABLE I
EXAMPLE RESULTS OF SEMINAR SESSION IN A CASE STUDY OF 16 STUDENTS

Student

Ss

Sp

P2P

Final

#1

Group
1

Si
8

0.4

Sf
5.6

6.96

#2

0.3

4.2

5.72

#3

9.5

0.4

7.4

7.64

#4

9.5

0.5

9.25

8.75

#5

9.5

0.5

9.25

9.15

#6

9.5

0.5

9.25

8.75

#7

0.3

4.2

5.32

#8

9.5

0.4

7.4

7.24

#9

0.3

4.2

5.72

#10

9.5

0.5

9.25

9.15

#11

9.5

0.4

7.4

8.04

#12

9.5

0.5

9.25

8.75

#13

9.5

0.5

9.25

9.15

#14

9.5

0.5

9.25

9.15

#15

9.5

0.4

7.4

8.04

#16

9.5

0.5

9.25

9.15

Table II compares the scores obtained from the seminar

groupal work, including the correction by the four
decorrelation techniques, with the individual score obtained in
the laboratory. This is a comparable measurement as the
practical work score is obtained via the on-line exam almost at
the same dates during the course progress.
TABLE II
EVALUATION RESULTS AND VARIANCE COMPARING WITH THE PRACTICAL
ONLINE EXAM SCORE

Student

Seminar Mean

Laboratory Mean

Variance

#1

7.25

7.9

0.20

#2

6.93

7.6

0.22

#3

7.65

8.3

0.20

#4

7.34

0.21

#5

8.02

8.6

0.16

#6

8.81

9.3

0.11

#7

5.5

5.8

0.045

#8

7.98

8.6

0.19

#9

5.88

6.5

0.18

#10

8.78

9.3

0.13

#11

7.98

8.6

0.19

#12

7.37

7.6

0.024

#13

8.95

9.6

0.21

#14

8.81

9.2

0.073

#15

6.98

7.6

0.18

#16

8.4

0.18

It can be observed in Table II that a maximum variance of

0.22 is obtained comparing the results of groupal seminar with
the practical on-line exam. This confirms the suitability of the
approach proposed.
The overall satisfaction of the alumni regarding groupal
interaction has been as measured by the Instituto de Ciencias
de la Educacin (ICE), an internal education-quality control
organism of the Universidad Politcnica de Valencia. The
obtained results of satisfaction of the alumni rose from 6.86
(over 10) to 8.24 points (over 10) in the academic year
2006/2007. This means a 20.12% increase compared with the
previous year.
V. CONCLUSIONS
Comparing the 2005/06 academic year, when this tool was
introduced, with the 2004/05 academic year, we observed a
9% increase in alumni passing the subject. After the
development of the data-mining tool, the introduction of the
control panel, and the associate corrective lecturing actions,
the pass-rate of the laboratory work in the subject has
increased in 37% the first year it was introduced. The
proposed interactive seminars comprised four phases, in
which the students work in small groups of up to five people
to solve a proposed exercise. The introduction of the groupal
seminar was reflected in an increase of 9.4% of the passing
alumni (from 61.2% to 70.6%) in the analyzed year
(2006/2007) over the first year of the curricula of the ASC
subject was set (2004/2005).
The results indicate good correspondence between the
individual scores obtained applying these techniques in the
laboratory and seminar lessons. A maximum variance of 0.22
is obtained. This indicates the suitability of the evaluation
techniques proposed for laboratory sessions and seminars.
It is also worth to note that the overall satisfaction of the
alumni regarding groupal interaction rose from 6.86 (over 10)
to 8.24 points (over 10) in the academic year 2006/07, i.e. a
20.12% increase, as measured by the Instituto de Ciencias de
la Educacin (ICE), an internal education-quality control
organism of the Universidad Politcnica de Valencia.
Further enhancements will be included in the herein
presented evaluation techniques in the next years targeting
optimize the data mining process considering alumni variables
collected in a several-years time frame.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]

The Bologna Declaration of 19 June 1999, Joint declaration of the

European Ministers of Education, European Higher Education Area,
http://ec.europa.eu/education/policies/educ/bologna/bologna_en.html
R. Zaane. "Web Usage Mining for a Better Web-Based Learning
Environment". University of Alberta, Canada, 2001.
I. H. Witten, E. Frank, Data Mining: Practical Machine Learning
Tools and Techniques, Ed. Morgan Kaufman, 2005.
C. Romero, S. Ventura, P. Espejo, C. Hervas. Data mining algorithms
to classify students. Educational Data Mining Conference EDM 2008.
Montreal, pp. 182-185, June 20-21, 2008.
K. A. Smith, Cooperative Learning: Effective Teamwork for
Engineering Classrooms, IEEE Frontiers in Education Conference,
Session 2b5, pp. 13-18, 1995.
W. Rau, B. S. Heyl, Humanizing the college classroom: Collaborative
learning and social organization among students. Teaching Sociology,
Vol. 18, pp.141-155, 1990.