Course Recommendation from Social Data

Hana Bydžovská

1,2

and Lubomír Popelínský

CVT, Faculty of Informatics, Masaryk University, Brno, Czech Republic

KD Lab, Faculty of Informatics, Masaryk University, Brno, Czech Republic

Keywords: Recommender System, Social Network Analysis, Data Mining, Prediction, University Information System.

Abstract: This paper focuses on recommendations of suitable courses for students. For a successful graduation, a

student needs to obtain a minimum number of credits that depends on the field of study. Mandatory and

selective courses are usually defined. Additionally, students can enrol in any optional course. Searching for

interesting and achievable courses is time-consuming because it depends on individual specializations and

interests. The aim of this research is to inspect different techniques how to recommend students such

courses. This paper brings results of experiments with three approaches of predicting student success. The

first one is based on mining study-related data and social network analysis. The second one explores only

average grades of students. The last one aims at subgroup discovery for which prediction may be more

reliable. Based on these findings we can recommend courses that students will pass with a high accuracy.

1 INTRODUCTION

Recommender systems aim to prioritise information

about items such as movies, music, books, news,

images or web pages to users with respect to their

interests. Jannach et al. (2011) presented different

types of recommendations. The selection is based on

the knowledge of user behaviour, information about

behaviour of other users, and information about of

all items in the database.

Recommender systems can be also used in an

educational environment. Students have to pass

many courses to finish their study. Some of them are

obligatory, but optional courses have to be chosen

by students. Students try to choose the best for

them–interesting and passable courses, but it is very

difficult to find suitable ones. Searching is very

time-consuming and students have to search whole

course catalogue, to examine abstracts and syllabi, to

check success rate statistics or ask other students for

their experiences.

To help students with their duties we intend to

design a course enrolment recommender system that

assists students when selecting courses. The

recommendation is based on educational data

mining and social network analysis methods. The

recommendation is personalized for each student.

The course enrolment recommendation can be

divided into two main parts: finding interesting

courses and checking if the courses are not difficult

for students. The second part is the most important.

When a student enrols in difficult course and fail,

the student can fail a study. The student would not

use such recommender system. Previous experiment

was published in Bydžovská et al. (2013).

This paper deals with recommendation of

courses that will not be too difficult for a particular

student. The aim of the proposed method is to

predict student success or failure in selected courses.

It is important not to recommend difficult courses

for particular students and it is equally important to

advise students about mandatory courses that usually

cause problems to students. We aim at identification

of such courses by using information that the

courses were problematic for students with similar

achievements.

The paper is organized as follows. In the

following section, we present related work. In

Section 3, we introduce the proposed recommender

system. In Sections 4 and 5 we describe used data

and in Section 6 we present experiments dealing

with predictions of study success. Results can be

found in Section 7. The discussion, summary and

future work can be found in the last two sections.

2 STATE OF THE ART

A recommender systems overview used in education

268

Bydžovská H. and Popelínský L..

Course Recommendation from Social Data.

DOI: 10.5220/0004840002680275

In Proceedings of the 6th International Conference on Computer Supported Education (CSEDU-2014), pages 268-275

ISBN: 978-989-758-020-8

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

can be found in Manouselis et al. (2011). A common

method to analyse educational data is to use

educational data mining methods (see Romero and

Ventura (2007)). It deals with the analysis of data

for understanding student behaviour. These

techniques can reveal useful information to teachers

and help them design or modify the structure of

courses. Students can also facilitate their studies

using the discovered knowledge. Nowadays,

researchers use educational data mining techniques

mostly to guide student learning efforts, develop or

refine student models, measure effects of individual

interventions or improve teaching support.

One of the most important issues often solved in

educational environment is understanding what

influences student performance. The task involves

the prediction of student's grades or student's course

difficulties. This information can identify students

with greater potential and also those that may

require timely help from teachers or peers to fare

well in the course.

Researchers usually mine from data stored in

university information systems. Mostly, they use

data such as grades, gender, field of study or age.

Thai Nghe et al. (2007) concluded that better results

were gained using decision trees than using

Bayesian networks.

Vialardi et al. (2009) aimed to select courses for

students in order to obtain good exam results.

Difficulties of courses were compared with student

potentials. Both variables were computed from

grades. The work extension can be found in Vialardi

et al. (2010) where the analysis was based on profile

similarity. The results were satisfactory but the false

positives obtained in results were too high. It is

worse to recommend a course that students enrol in

and fail than missing a course that they could pass.

The solution was to sample the data again. It

lowered the accuracy, but decreased significantly the

false positive errors.

Another common topic of mining in educational

data is the prediction of drop-out rate of students.

Dekker et al. (2009) explored the possibilities of the

assignment. The task is similar to the student's

performance analysis but we are interested in the

complex performance and in the chance to

successfully complete their studies .

Our previous work also explored drop-out

prediction (Bayer et al. (2012)). We collected useful

information about students’ studies. We applied

educational data mining methods to this data. We

then created a sociogram from the social data. We

used social network analysis methods to this data

and obtained new attributes such as centrality,

degree or popularity, etc. When we enriched the

original study-related data with these social

attributes and employed educational data mining

methods again, the accuracy of classification

increased from 82.5% to 93.7%.

Marquez-Vera et al. (2011) used questionnaires

to get some detailed information of students’ lives

directly from students because this type of data is

not present in the information system, e.g. the family

size, the smoking habits or the time spent doing

exercises. These data can improve predictions about

students failure.

In this work, we applied data mining methods to

explore the study-related data. Unlike Marquez-Vera

(2011) who was dependent on answers from a

questionnaire, we used confirmed and complete data

from the university information system. If compared

with Thai Nghe et al. (2007) we tested broader

spectrum of machine learning algorithms—bayesian,

as well as instance-based learners, decision tree and

also various rule-based learners. We further

extended the method of Vialardi et al. (2009) by

addition of social data. In this way we were able to

compare students' data together with the information

about their friends. Therefore, we could increase

prediction accuracy.

3 A RECOMMENDER SYSTEM

PROPOSAL

Students are interested in information resources and

learning tasks that would improve their skills and

knowledge. The recommender system should, hence,

monitor their duties and show them either an easy or

an interesting way to graduate.

The proposal of recommender system consists of

three parts: data extraction module that extracts data

from the Information System of Masaryk University

(IS MU) database, pre-processing and analytical part

(allows the user to select relevant features, to

compute new ones, to obtain basic statistics about

those features, and to run machine learning

algorithms) and the presentation module (selects

important knowledge and presents it to the user).

3.1 Use of the System

The proposed system will recommend mandatory

courses and associated prerequisite courses. Elective

and optional courses will be selected according to

the student's potential with respect to vacancies in

the timetable. The system will recommend

interesting, beneficial and achievable courses for

CourseRecommendationfromSocialData

269

clever students. On the other hand, for weak students

it will search for courses that can contribute

knowledge to finish mandatory and elective courses.

Passing all mandatory and elective courses

guarantees that a student deserves a university

degree. When the system finds a difficult mandatory

course for a student, it can inform him or her about

the situation and the student can pay attention to the

course and study hard. When a student needs to

select elective or optional courses for a term, the

recommender system selects interesting, but

passable courses for a particular student.

The system will eventually recommend

interesting and passable courses to students and will

propose a short explanation of its decision and

confidence. Students will have an opportunity to

assess each recommendation if recommended

courses were interesting and adequate difficult.

Based on the assessments, recommendation

algorithms will be modified to enhance the relevance

of recommendations. The recommendations will be

available for students of Masaryk University

probably from autumn 2014

4 SOCIAL AND STUDY-RELATED

DATA EXTRACTION

Selecting attributes that express student’s

characteristics as accurately as possible is extremely

important. Based on such data, we can give a better

prediction on the courses that are crucial or

interesting for a student. We tried to obtain such

attributes that tell us as much as possible about

students and their lives. The list of all attributes can

be found in Section 5.1.

We believe that schoolmates who become friends

have much in common. Although we cannot find it

in the data, they can have similar sense of humour,

close interests and maybe same intellect capability to

be able to spend time and enjoy together. It is so far

hypothetical, but very likely, that students with

clever friends will have better study results than

students with the same potential who do not have

such friends. To observe this, we explore social ties

among students.

4.1 Social Behaviour Features

There are a number of interpersonal ties that have

been already evaluated to enhance IS MU full text

search. Some ties are intuitive: (a) explicitly

expressed friendship, (b) mutual email conversation,

another person. Weaker ties are more hidden and are

derived from the following facts: (e) discussion

forum message marked as important, (f) whole

thread in discussion forum or blog marked as

favourite, (g) files uploaded into someone else's

depository, (h) assessments of notice board's

messages, (i) visited personal pages.

We measured the value of a tie by its importance

and weighted by a number of occurrences. As a

result we calculated a single number from all

mentioned ties reflecting the overall strength of

student's relation with any given schoolmate.

A sociogram, a diagram which maps the

structure of interpersonal relations has been created

from information about students, their direct friends

and relations among them. This allow us to compute

new student features from the network structural

characteristics and student direct neighbours

attributes using tools for social network analysis,

e.g. Pajek. These features give us a new insight into

the data. The list of computed social behaviour

attributes can be found in section 5.2.

5 DATA

We use three types of data: study-related data, social

behaviour data and data about previously passed

courses.

5.1 Study-related Data

This type of data represents student and his or her

achievements.

Personal attributes: (a) gender, (b) year of birth,

score—a result of the entrance examination

expressed as the percentage of the score measuring

learning potential—minimum of all attempts to get

at the university.

Historical attributes (include all student's

outcomes achieved before the term in which the

student attended the investigated course): (e) credits

to gain—a number of credits to gain for enrolled, but

not yet completed courses, (f) gained credits—a

number of credits gained from completed courses,

(g) a ratio of the number of gained credits to the

number of credits to gain, (h) courses not

completed—a number of courses a student has failed

to complete, (i) second resits done—a number of

used second resits (an examination taken by a

previously unsuccessful student), (j) excused

days—a number of days when a student is excused,

CSEDU2014-6thInternationalConferenceonComputerSupportedEducation

270

(k) average grades—an average grade computed

from all grades obtained, (l) weighted average

grades—average grades weighted by the number of

credits gained for courses.

Term-related attributes (information about a term

and a study in which the student enrolled in the

investigated course): (m) field of study, (n) program

of study, (o) type of study (bachelor or master), (p) a

number of terms completed, (q) a number of parallel

studies at the faculty, (r) a number of parallel studies

at the university, (s) a number of all studies at the

faculty, (t) a number of all studies at the university.

5.2 Social Behaviour Data

We computed social attributes for each student from

sociogram we described in section 4.1: (a) degree—

represents how many relations the student is

involved in, (b) weighted degree—degree with

respect to strength of the ties, (c) closeness

centrality—represents how close a student is to all

other students in the network, (d) betweenness

centrality—represents student's importance in the

network, (e) grade average of neighbours—

calculation of average grades of the nearest

neighbourhood values, (f) neighbours count in

course—how many nearest neighbours have already

enrolled in the course.

In our interpretation, the degree measures the

amount of communication of each student. The

closeness centrality measures distances needed to

get some information from a student to all other

students in the sociogram. The betweenness

centrality expresses the frequency of a student in the

information path between two different students.

5.3 Courses Passed by a Student

We added this type of data because we believed that

the knowledge of passed courses is important and

influences student performance. This type of data

contained all passed courses for each student in the

data set. We used only information about passing or

failure in these experiments, we were not interested

in exact grade because we observed that an exact

grade is not important.

5.4 Data Sets

For exploring course difficulties we chose some

courses of Masaryk University:

 IB101 Introduction to Logic

 IA008 Computational Logic

 IB108 Algorithms and data structures II

 IA101 Algorithmics for Hard Problems

 MB103 Continuous models & statistics

These courses are offered mainly for students of

Applied Informatics, one of the programmes in the

Faculty of Informatics. The choice was made with

respect to importance of courses to students, how

courses relate to one another, and the lecturers for

the courses.

We generated two data sets for each of the

above-mentioned courses. We used data from the

years 2010-2012. As we aimed at predicting student

success from historical data, the years 2010 and

2011 were used for learning. A test set then

contained data about students who attended a

particular course in the year 2012. A number of

instances in the data sets is presented in Table 1.

Table 1: Number of instances.

Course Data sets No. of

students

No. of vertices

in sociogram

IB101 Training set 782 24829

Test set 427 16649

IA008 Training set 158 6808

Test set 73 5713

IB108 Training set 127 10652

Test set 56 6335

IA101 Training set 219 11338

Test set 113 9505

MB103 Training set 708 24018

Test set 331 14495

6 METHODS

A recommender system core is an analytical module

that exploits various machine learning algorithms

from Weka (see Witten et al., 2011). The current

version of the module contains three methods that

comprise recommendation from complete historical

data then learning based on grade averages, and also

discovery of student subgroups for which a

recommendation may be more promising. An

obtained accuracy was always compared with a

baseline, i.e. with the accuracy when all the data in a

test set were classified into a majority class.

CourseRecommendationfromSocialData

271

6.1 Mining Complete Data

The first method aims at classification of student's

ability to pass an investigated course. We tested

different machine learning algorithms—naive Bayes

(NB), Support Vector Machines (SMO), instance-

based learning (IB1), two rule learners (PART and

OneR), decision tree (J48) and two ensemble

learners (AdaBoost (AdaB) and Bagging).

Three experiments were performed that differ in

granularity of a class—prediction of an exact grade

A-F, prediction into three classes: good/bad/failure

and two-class prediction of success/failure. We used

three collections of attributes for classification: All

data (study-related attributes together with social

behaviour data), only study-related data (all study-

related data without social behaviour data), subset of

attributes (the best subset of attributes selected by

feature selection algorithms—GainRatioAttEval,

InfoGainAttributeEval and CfsSubsetEval). We also

enriched all of the collections with information

about students' previously passed courses.

6.2 Comparison of Grade Averages

The second method inspired by Vialardi et al. (2009)

was based on a comparison of average grades of a

student with average grades for the investigated

course. The designed method also considered grades

of students' friends. We computed the average grade

from training set for all courses and predicted the

study performance in the test set. The course average

grade was compared with the student's potential,

which was measured as follows: (a) average of

student grades, (b) average of all student's friends'

averages from the sociogram, (c) average of

averages of student's friends that attended the

investigated course simultaneously with the student.

If the course average grade was higher than the

student's potential, we predicted success and failure

otherwise.

6.3 Recommendations to Subgroups

For subgroup discovery (see Lavrač et al., 2002,

2006) we combined discovery of finding interesting

subsets of attribute values (by means of

discretization for continues attributes and by

building subsets of values for categorical attributes)

with two learning algorithms—decision trees (J48)

and class association rules (see Liu et al., 1998,

Witten et al., 2011).

We first computed subsets of values for each

attribute—from 5 to 10 bins in case of discretization,

and couples and triples for categorical attributes—on

the learning set. For each combination of such

attributes we then learned decision rules extracted

from decision tree (see Quinlan, 1993) and class

association rules. From all rules with coverage

higher than 5% of test set cardinality we choose

those that had precision at least 5% higher than the

best precision reached in the previous experiments.

7 RESULTS

The aim of these experiments was to recommend a

course to a student based on the analysis of historical

data. Some students rely on getting really good

grades and not only on passing successfully, which

is why we attempt to predict an exact grade and

subsequently, either recommend a course or to warn

a student not to enrol in the course. If the system

recommended a course that is hard to pass or even

non-passable for a student, the recommendations

would not meet expectations.

7.1 Mining Complete Data

The results of the first experiment—classification

into classes according to grades A, B, C, D, E, F

(Table 2)—are not too convincing and also the

accuracy improvement is quite small when

compared with the baseline. It supports the

observation that there is no strong difference

between students when the difference in grades is

small.

The obtained results of three class classification:

good/bad/failure (Table 3)

yield higher accuracy

than the previous one. The maximum difference

from baseline was observed for IB108—18%. If

compared to Bydžovská et al. (2013), accuracy

increased for 4 out of 5 courses. Only exception was

MB103 where the accuracy remained unchanged.

Table 2: Classification into classes according to grades.

Course Baseline Data Best results

IB101 40.74% Subset + Courses 43.33% AdaB

IA008 34.24% Subset 39.72% J48

IB108

17.86% Study-related data 33.92% PART

Subset 33.92% IB1

IA101 38.93% All data 42.47% SMO

MB103 28.09% Subset + Courses 32.63% Bagging

CSEDU2014-6thInternationalConferenceonComputerSupportedEducation

272

Table 3: Three class classification: good/bad/failure.

Course Baseline Data Best results

IB101 68.38% Subset + Courses 68.62% AdaB

IA008 56.16% Subset + Courses 66.67% SMO

IB108 44.64% Subset + Courses 62.50% NB

IA101 53.09% Subset + Courses 65.49% AdaB

MB103 47.12% Study-related data 57.70% Bagging

As we could see in results above, for grade

prediction none of classifiers was able to reach

accuracy significantly higher than baseline. For

classification of success or failure (Table 4), the case

was different. For success/failure prediction, for all

of subjects, but IB101 there was slight improvement

in accuracy. For IB108 the accuracy reached 82.14%

what was more than 10% increase. Even higher

increase—more than 25%—was observed for

IA101. Data about students' previously passed

courses improved the results in this case.

Table 4: Classification of success or failure.

Course Baseline Data Best results

IB101 91.10% Subset 90.16% SMO

IA008 83.56% All data 89.04% SMO

IB108 69.64% Study-related data 82.14% SMO

IA101 53.10% All data +

Courses

81.42% AdaB

MB103 69.48%

Study-related data 75.22%

NB/Bagging

7.2 Comparison of Grade Averages

This method, as introduced in 6.2, was based on

comparison of average grades of the student with

average grades for the investigated course. In

Table 5, (a) contains results when the student grade

was compared with average grades of other students,

with average of all student's friends' averages from

the sociogram (b), and average of averages of

student's friends that attended the investigated

course simultaneously with the student (c).

This method resulted in slight accuracy increase

in most cases for the choice (b)—average of all

student's friends' averages from the sociogram. All

results can be seen in Table 5.

Based on those results, we decided to build an

ensemble learner that employs those three

classifiers. A course is recommended to a student

only if all three classifiers predict success. In the

same manner, the course is not recommended if all

three classifiers predict failure. Otherwise, the

classifiers do not supply any recommendation.

Table 5: Prediction of student success from student

potential.

Course Baseline (a) (b) (c)

IB101 91.10% 50.58% 91.29% 75.00%

IA008 83.56% 59.72% 84.28% 84.84%

IB108 69.64% 64.28% 70.90% 61.11%

IA101 53.10% 61.94% 46.90% 54.63%

MB103 69.48% 63.74% 69.48% 67.28%

The results in Table 6 show significant

importance of social ties between students. It

supports hypothesis that students having clever

friends have higher probability to pass courses than

the others.

Table 6: Ensemble learner of student potential.

Course Successful

students

Predicted to

be successful

Precision Recall

IB101 390 167 98.80% 42.30%

IA008 60 36 91.67% 55.00%

IB108 39 24 87.50% 53.84%

IA101 53 78 56.41% 83.01%

MB103 230 123 92.68% 49.56%

7.3 Recommendations to Subgroups

In this experiment we looked for subgroups with

high precision of recommendations. The most

promising attributes were: the average grade and the

ratio of a number of gained credits to a number of

credits to gain (credits ratio). The best results for

each course are in Table 7.

Table 7: Discovered subgroups.

Course Attribute Range Precision Recall

IB101 Avg. grade (-inf, 1.8> 98.60% 8.95%

IB108 Credits ratio (-inf, 1.20> 85.56% 81.10%

IA101 Credits ratio (-inf, 0.23> 77.40% 17.35%

MB103 Credits ratio (-inf, 1.29> 96.43% 49.15%

We also explored manual invention of subgroups.

We focused on the field of study and the year when

CourseRecommendationfromSocialData

273

the exam was passed. We observed that the accuracy

increased between 2 and 4% for the field of study.

However, this approach needs to be further

elaborated.

8 DISCUSSION

We observed that use of social data together with

study-related data resulted in accuracy increase in

most of cases. On the other side, when using only

social behaviour data, results were worse than when

using only study-related data.

The most useful attributes were almost all social

behaviour attributes—closeness centrality, both

types of degree and betweenness centrality. The

most promising attribute was closeness centrality.

We may conclude that the most important is how

fast a student can get a certain information from

other students in the sociogram. Among study-

related attributes it was an average of grades, a

weighted average of grades, credits to gain, gained

credits, a programme and a field of study.

The results were also improved by adding the

information about student previously passed courses.

The largest improvement was observed at course

IA101. It may be caused by the fact that students

usually enrolled in this course later than in the other

courses that were included in this research.

The next observation concerns ensemble learner

of student potential (Table 6 in 7.2). The learner

significantly improved precision if compared with

experiment from 7.1. The price is lower recall we

are capable to give right recommendation only to a

subpart (about 50%) of students. Concerning

subgroup discovery, results for IA101 and MB103

were improved but we did not succeed in

discovering an interesting subgroup for IA008. It

may be also useful to combine the first two

methods—machine learning and average grade

comparison—and apply such an ensemble learner to

promising subgroups of students.

We observed that experimental results were

worse for courses that changed in the period of

2010-2012. That change may concern contents of

the course or a way in which students have been

evaluated. In that case learning and test data may not

be from the same distribution what usually causes a

decrease of performance, i.e. accuracy. To prevent

from such a situation it would be necessary to check

compatibility of historical (training) data and

current (test) data e.g. by the methods described in

Jurečková et al. (2012).

9 CONCLUSIONS AND FUTURE

WORK

Our main contribution is to provide a method to use

social data together with other educational data for

course prediction. We presented three different

methods to recognize and recommend passable

courses to a student and warn against difficult ones.

The proposed methods were validated on

educational data originated in IS MU. We used

different analytical tools, namely machine learning

algorithms, comparison of student grade averages

and employed also subgroup discovery. We

concluded that for most of courses we could provide

a recommendation to students.

There is still room for future improvements.

Some of recommendations suffer from low

confidence. In the future work we will use more

detailed history of study. We also plan to introduce

temporal attributes and to employ algorithms for

mining frequent temporal patterns. We plan to

extend data with time stamps (e.g. about the term in

which a student passed a course) and to employ

sequence pattern mining because the time sequence

in which a student passed courses can be beneficial.

The information system also contains data about

online tests that a student passed and also

information student access to online study materials.

Such statistics enabled us to better understand

student learning habits. Students learning

continuously should be more successful than the

others. We also intend to use the timetable data of

course lessons. Some students can have problems

with morning or late afternoon lessons and it can

influence the course final grade. This information

could enrich student characteristics and improve

prediction. We can also enrich the data with

information obtained from Course Opinion Poll

where students evaluate courses, use similarity

algorithms and predict the difficulty of the

investigated course for a particular student based on

the similarity of responds with others. We can

compare our predictions with a student’s subjective

opinion about courses they have already passed and

with results from similarity experiments.

Whenever a system will be running (we suppose

that this autumn term is a realistic estimate) a

student feedback will be the most important source

of information.

ACKNOWLEDGEMENTS

We thank Michal Brandejs, IS MU development

CSEDU2014-6thInternationalConferenceonComputerSupportedEducation

274

team colleagues and Knowledge Discovery Lab

members for their assistance. We also thank Alex

Popa for his help. This work has been partially

supported by Faculty of Informatics, Masaryk

University.

REFERENCES

Bayer, J. and Bydžovská, H. and Géryk, J. and Obšívač, T.

and Popelínský, L. (2012) Predicting drop-out from

social behaviour of students. In Kalina Yacef et al.

(eds.). Proceedings of the 5th International

Conference on Educational Data Mining - EDM 2012.

Greece: www.educationaldatamining.org, pp. 103 -

109, ISBN 978-1-74210-276-4.

Bydžovská, H. and Popelínský L. (2013) Predicting

Student Performance in Higher Education. In Franck

Morvan et al. (eds.) 24th International Workshop on

Database and Expert Systems Applications - Dexa

2013. IEEE Computer Society, 2013. pp. 141-145, 5.

Dekker, G.W. and Pechenizkiy, M. and Vleeshouwers,

J.M. (2009) Predicting students drop out: a case study.

In T. Barnes et al. (eds.), Proceedings of the 2nd

International Conference on Educational Data Mining

(EDM’09), pages 41–50.

Jannach, D. and Zanker, M. and Felfernig, A. and

Friedrich, G. (2011) Recommender Systems An

Introduction, Cambridge University Press.

Jurečková, J. and Kalina J. (2012) Nonparametric

multivariate rank tests and their unbiasedness.

Bernoulli 18(1), pp. 229–251, DOI: 10.3150/10-

BEJ326.

Lavrač, N. and Železný, F. and Flach, P. A. (2002) RSD:

Relational Subgroup Discovery through First-Order

Feature Construction. In 12th International

Conference on Inductive Logic Programming, pp.

149—165.

Lavrač, N and Kavšek, B (2006) APRIORI-SD: Adapting

Association Rule Learning to Subgroup Discovery.

Applied Artificial Intelligence, pp. 543-583, ISSN:

0883-9514.

Liu, B. And Hsu, W. And Ma, H. (1998) Integrating

classification and association rule mining. Proceedings

of the 4th international conference on Knowledge

Discovery and Data mining KDD'98, page 80--86.

AAAI Press.

Manouselis, N. and Drachsler, H. and Vuorikari, R. and

Hummel, H. and Koper, R. (2011) Recommender

Systems in Technology Enhanced Learning,

Recommender systems Handbook Spriger Verlag

2011, pp 387-415.

Marquez-Vera, C. and Romero, C. and Ventura, S. (2011)

Predicting school failure using data mining. In Mykola

Pechenizkiy et al. (eds.), EDM, pages 271–276.

www.educationaldatamining.org. ISBN 978-90-386-

2537-9.

Quinlan, J. R. (1993) C4.5: programs for machine

learning. Morgan Kaufmann Publishers Inc. San

Francisco, ISBN:1-55860-238-0.

Romero, C. and Ventura, S. (2007) Educational data

mining: A survey from 1995 to 2005. Expert Syst.

Appl., 33(1):135–146, ISSN 0957-4174.

Thai Nghe, N. and Janecek, P. and Haddawy, P. (2007) A

comparative analysis of techniques for predicting

academic performance, 37th ASEE/IEEE Frontiers in

Education Conference, Milwaukee, WI.

Vialardi, C. and Agapito, J. and Shafti, L. and Ortigosa, A.

(2009) Recommendation in higher education using

data mining techniques. In T. Barnes, M. Desmarais,

C. Romero, and S. Ventura (eds.), Proceedings of the

2nd International Conference on Educational Data

Mining (EDM’09), pages 191–199.

Vialardi, C. and Chue, J. and Barrientos, A. and Victoria,

D. and Estrella, J. and Ortigosa, A. and Peche, J.

(2010) A case study: Data mining applied to student

enrollment. In Proceedings of Third Educational Data

Mining Conference, Pennsylvania, USA, pages 333–

335.

Witten, I. and Frank, E and Hall, M. (2011) Data Mining:

Practical Machine Learning Tools and Techniques, 3

edition, Morgan Kaufmann Publishers (ISBN: 978-0-

12-374856-0).

CourseRecommendationfromSocialData

275