IEEE ACCESS Final Recommender System
IEEE ACCESS Final Recommender System
net/publication/378458769
CITATION READS
1 442
5 authors, including:
Khondaker A. Mamun
University of Toronto
106 PUBLICATIONS 735 CITATIONS
SEE PROFILE
All content following this page was uploaded by Arifur Rahman on 12 March 2024.
ABSTRACT Recommender systems have gained significant attention as powerful tools for supporting
decision-making processes in various domains. However, the understanding of their impact and application
in the field of academic choices in higher education remains limited. This systematic review aims to provide
a comprehensive summary of the current knowledge regarding recommender systems utilized in the context
of academic choices and advising in higher education. The study is based on the systematic analysis of
a set of primary studies (N = 56 out of 1578, published between 2011 and 2023) included according to
defined criteria. The articles were categorized based on specific criteria, and their findings were analyzed
and synthesized. Results show that the hybrid strategy has been the most effective method for producing
recommendations. Evaluation measures such as offline experiments and case-study validation were promi-
nently observed in the empirical studies, providing insights into the effectiveness of recommender systems.
The findings reveal that the design of recommender systems in higher education is context-specific, with
researchers considering various parameters to tailor recommendations to individual needs. However, most
of the selected articles relied on lab-based studies rather than real-world applications, indicating a need
for further research in practical settings. This systematic review also identifies future research directions,
including the incorporation of deep learning technologies and the analysis of personality traits. By providing
a comprehensive overview of the current state of recommender systems for academic choices in higher
education, this review offers valuable insights for researchers and practitioners, guiding the development of
more effective and personalized recommendation systems to unlock the full potential of individuals in their
academic journey.
INDEX TERMS Academic choices, higher education, recommendation systems, course recommendation
systems, holland code assessment, systematic literature review.
limited attention. Understanding how recommender ficulties arise from differences in learner’s edu-
systems can support students’ academic decision- cational interests and needs [8]. These challenges
making processes is crucial for providing effec- include data sparsity, cold start [7], privacy con-
tive guidance and enhancing educational outcomes. cerns, ensuring diversity in recommendations, and
Recommender algorithms can be used to make addressing biases that may arise from the data used
smart decisions in complex information systems and to train the system. The approach utilized to produce
help the users decide upon useful materials [5]. The recommendations is also an example of difficulties.
purpose of this systematic review is to provide a For instance, the way content-based recommender
comprehensive summary of the current knowledge systems manage data is inextricably linked to over-
regarding the utilization of recommender systems specialization [9], [10]. Accurately identifying user
in the context of academic choices and advising expectations and recommendations is one of the
in higher education. The review categorizes the main challenges [11]. Differences in learners’ ed-
selected articles based on specific criteria, allowing ucational preferences and needs lead to challenges
for a structured analysis of the various dimensions [8]. There are questions about how to assess the
and approaches adopted in the design and imple- effectiveness of RS from an educational perspec-
mentation of recommender systems for academic tive. Applying the classic recommender evaluation
choices. Notably, the articles primarily focus on methodologies is a common strategy to assess the
three distinct domains of academic choices, with quality of educational recommenders [12]. This
course recommendation being the most prevalent method evaluates the performance characteristics
area of interest. of the system, such as its precision and prediction
Students’ often get confused while choosing the accuracy. However, system effectiveness in the ed-
most suitable course in higher education that meets ucational setting must take into account students’
their requirements. This confusion arises due to the learning progress. This feature adds significant chal-
vast array of available courses and the complexity lenges to how to assess RS effectively from an
of educational systems. To address this challenge, educational perspective. The scientific community
recommender systems have emerged as valuable has become increasingly interested in RS [13], and
tools for supporting students in making informed in recent years, substantial study has been done
decisions about their academic choices in higher to solve these concerns [6], [14], [15], [16]. Data
education. Recommender systems utilize advanced mining, information filtering, education and infor-
algorithms and techniques, such as machine learn- mation technologies, machine learning, and other
ing and data mining [6], to analyze relevant data computational approaches are only a few examples
and generate personalized recommendations for stu- of how RS has evolved into an area of application
dents. By considering factors such as students’ pref- [6] in education [17].
erences, academic performance, career aspirations, Despite these challenges, recommender systems
and feedback from previous students, these systems hold great potential to revolutionize the course se-
can assist in identifying the most suitable courses lection process in higher education. The efficient
that align with individual requirements. The benefits design of the recommender system in education
of recommender systems in academic choices are will help the students by generating the appropriate
manifold. Firstly, they save students time and effort recommendations [18]. As technology advances and
by providing them with tailored recommendations, more data becomes available, it is expected that
eliminating the need for extensive manual research. recommender systems will become even more accu-
Secondly, they enhance the accuracy of decision- rate and effective in providing personalized recom-
making processes by considering a wide range of mendations to students. In addition, recommender
relevant information and filtering out irrelevant op- systems offer valuable support to students facing the
tions. Moreover, these systems promote personal- daunting task of choosing the most suitable courses
ized learning experiences, enabling students to ex- in higher education. By leveraging advanced algo-
plore courses that align with their interests and rithms and analyzing relevant data, these systems
goals, Recommender System (RS) can contribute can alleviate confusion, save time, and enhance
to their academic performance and motivation by the accuracy of decision-making. As research and
indicating personalized learning content [7]. development in this field continue to progress, rec-
However, there are challenges associated with the ommender systems will play an increasingly vi-
development and implementation of recommender tal role in helping students navigate the multitude
systems in the context of academic choices. Dif- of course options [19], ensuring they make well-
2 VOLUME 11, 2023
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
TABLE 1: Research questions that are the focus of this systematic review
informed choices that align with their academic and II. LITERATURE REVIEW
professional aspirations. Recommender systems (RSs) have gained signifi-
This review article is organized as follows: In Sec- cant interest in education as technologies support-
tion II, related works are presented. Section III ing personalized teaching and learning experiences.
describes the protocol applied to conduct the sys- Over the past 12 years, research efforts have focused
tematic literature review. Section IV presents a syn- on mapping and summarizing various aspects of
thesis of the important results guided by research RSs in education. By exploring RSs in education,
questions. Section V highlights the comprehensive researchers aim to enhance teaching and learning by
outcome of this review and identifies some research tailoring educational content to individual learners.
gaps and possible solutions. Section VI describes The scientific community’s attention to RSs reflects
future research direction in this field. Section VII a growing interest in utilizing technology to improve
represents the practical uses of course recommen- educational experiences through personalized rec-
dation system. Finally, Section VIII concludes this ommendations.
review. In [20] a comprehensive review of technology-
enhanced learning recommender systems was car-
ried out. The authors analyzed 82 recommender
VOLUME 11, 2023 3
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
systems from 35 different countries published from ular research questions [24], [25]. Although con-
2000 to 2014 and provided an overview of the area. ventional reviews aim to summarize the findings
This study explores various aspects of recommender of several studies, systematic reviews use precise
methods, information sources, and assessments in and strict guidelines to select, critically analyze,
education and information technologies. It catego- and summarize all research on a specific topic. We
rizes selected publications using a provided frame- have performed a systematic literature review by
work. considering the methods outlined in [24] to obtain
In [21], Rivera et al. conducted a systematic a synopsis of recommendation systems regarding
mapping to present a comprehensive overview of higher educational academic choices. Our pipeline
the ERS (Educational Recommender Systems) do- involves several stages to conduct the systematic
main. Their study covered a broad range of pa- literature review, divided into three phases, each
pers and aimed to identify global characteristics in producing an output [25] and summarized in Fig.
ERS research. Similarly, Pinho, Barwaldt, Espín- 2. Details concerning the steps of our systematic
dola, Torres, Pias, Topin, Borba, and Oliveira (2019) review are described in the following subsections:
performed a systematic review of ERS, focusing
on different questions and utilizing different repos- A. REVIEW PLANNING
itories. Both works shared a common concern in The first phase involves defining the underlying
providing insights into the evaluation methods of need for systematic review, formulating research
these systems and the main techniques employed in questions (RQs) to guide the work, developing
the recommendation process. the research protocol, which involves identifying
In their research, [22] focused on course recom- the keywords and operators, exploring literature
mendation systems and conducted a comprehensive sources, defining inclusion and exclusion criteria,
review of techniques and parameters used in this and establishing a search strategy.
type of Educational Recommender Systems (ERS).
Additionally, they defined a taxonomy of the factors 1) Search, Selection, and Extraction
considered in the course recommendation process. The second phase includes searching for and identi-
On the other hand, in [23], the authors conducted fying relevant research, selecting research articles,
a review on affectivity based ERS. Their study analyzing quality by applying criteria, retrieving
presented a macro analysis, identifying key authors and monitoring data, synthesizing observations, and
and research trends while summarizing various as- identifying possibilities for future research.
pects of recommender systems related to affectivity.
These aspects included the techniques employed in B. RESEARCH QUESTION
affectivity analysis, the sources of affectivity data Formulating the research question(s) (RQ) is the
collection, and the methods used to model emotions. critical component of any systematic review, as it
In [16] the authors present a systematic literature focuses the study, determines the methodology, and
review on recommender systems in the educational guides all the stages of analysis and reporting [24].
domain. They analyzed 16 out of 756 primary Our research purpose is to examine an overview of
studies, published from 2015 to 2020. The review the related work in using recommender systems re-
reveals the dominance of the hybrid approach for garding higher educational course advising systems.
recommendation production and the limited focus Therefore, to achieve a better understanding of the
on accuracy in evaluation studies. It emphasizes the current literature, it is vital to formulate a series
need for multidimensional evaluation frameworks of research questions, each one addressing various
to assess the pedagogical effectiveness of recom- facets of RS in higher educational choices. Table
menders. The paper also identifies and discusses key I outlines all the research questions of this review
limitations, highlighting areas for future research to along with their rationale.
enhance recommender systems in education.
C. SEARCH STRATEGY
III. METHODOLOGY In this systematic literature review, we employed a
A systematic literature review (SLR) primarily in- well-structured search strategy in two stages. Firstly,
cludes a comprehensive and rigorous plan and we identified digital repositories for searches, in-
search strategy to minimize bias by defining, as- cluding ACM Digital Library, IEEE Xplore, Science
sessing, and categorizing all related research studies Direct, and Semantic Scholar, conducted through
on a specific topic, providing answers to partic- Google Scholar. Secondly, we defined keywords and
4 VOLUME 11, 2023
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
FIGURE 1: Detailed workflow of the systematic search, based on the PRISMA workflow guidelines
combined them with Boolean operators to create D. SELECTION CRITERIA & SCREENING
search strings. The list of keywords and correspond- PROCESS
ing search strings used for each digital library is To ensure that selected research articles meet the
presented in Table II and Table III, respectively. The scope of this systematic review, we applied an
search resulted in an initial set of 1578 potential inclusion-exclusion criterion to the initially ob-
primary studies for assessment of eligibility and tained studies to determine whether a paper should
inclusion. be included in the final review. Therefore, the in-
clusion and exclusion criteria are defined as Table
2. We performed preliminary selection, defined the
inclusion and exclusion criteria, and appraised the
initially selected papers. Paper titles, abstracts, and
VOLUME 11, 2023 5
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
to categorize mandatory and elective courses. It cal- single score denoting the degree of suitability of a
culated the effect rate between courses and extracted course for target students using Backpropagation.
rules using the C5.0 decision tree, finally developing Furthermore, [60] utilizes an ontology-based hybrid
a fuzzy logic model based on these rules. filtering framework comprising CB and CBF filter-
In article [30], the authors combine a hybrid ing, with CBF measuring similarities between user
approach of multiclass Support Vector Machine preferences and courses, and CF observing user-
(SVM) and K-Nearest Neighbor (KNN) to rec- user similarities in the system. The recommendation
ommend appealing graduate programs. A. Ragab list is then generated by combining the scores of CF
proposes in [34] and [35] a recommender system & CBF. Besides, model-based CF is another choice
comprising two cascading recommenders and a pre- for implementing the RS framework. [27] applies a
dictor to provide university recommendations and CF RS using K-means clustering to cluster similar
predictions. Both track and college recommenders students and find similarity between target students
use knowledge discovery rules, while the predictor and cluster groups by utilizing the N-nearest neigh-
compares current and previous student data avail- borhood technique, and generates a list of elective
able in the system, forecasting the most fitting col- course recommendations along with the expected
lege for a student. grade by applying the association rule mining algo-
Another choice of hybridization technique in- rithm. [66] aims to provide a 4-year study strategy
volves combining collaborative filtering (CF) and by considering multiple constraints. The authors
content-based filtering (CBF). In [40], CF and CBF propose a hybrid model in which the min-cost-max-
techniques are combined, where CF computes user flow algorithm is used to solve credit constraints
similarities using Euclidean distance, and CBF cal- and evaluate the usefulness of courses by combin-
culates student interest in academic tracks to rec- ing the scores of Course registration possibility by
ommend a suitable diploma track. [26] presents a matrix factorization, student performance prediction
two-stage collaborative filtering approach, employ- by CBF, career interest ratings, and interest levels on
ing an Artificial Immune System to predict course skills thus generating a learning plan by sorting a list
grades and make recommendations. [49] combines of courses into a directed graph and determining the
Alternating Least Square (ALS), a model-based CF priorities of courses.
algorithm, with TF*IDF, a popular content-based
filtering approach, to recommend suitable courses 2) Collaborative Filtering
for college students. In [58], personalized elective Collaborative filtering is a traditional filtering ap-
course recommendations are proposed, with two proach for information filtering, which represents
preference estimations calculated based on student 20% of the total of the studies. Paper [36] rep-
and course information separately, and a genetic resents Multi-Criteria Collaborative Filtering (MC-
algorithm configures the relevance of each crite- CF) along with Dimensionality Reduction tech-
rion to provide user-specific suggestions. In [59], niques to yield university or college recommen-
a course recommendation system measures simi- dations. [47] proposes a web-based course advis-
larity of course topics, matches descriptions and ing system using model-based CF (K-means algo-
tags between courses and users’ profiles and level, rithm), whereas study [48] represents a framework
and analyzes course sentiments using Linear dis- for university elective courses utilizing Pearson Re-
criminant analysis (LDA), word-correlation factors, lationship Coefficient and Alternating Least Square
and SentiWordNet score respectively. It predicts (ALS). The authors in [50] also recommend a course
ratings using Matrix Factorization and generates a with CF based Bayesian Personal Ranking Matrix
8 VOLUME 11, 2023
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
Factorization (BPRMF) algorithm, whereas the au- behaviors, followed by converting variables into
thors in [52] recommend suitable study programs fuzzy variables and mining fuzzy association rules
using item-based CF algorithms. [54] outlines a in each cluster, yielding course selection rules to
design of a recommendation system based on the recommend courses along with the predicted scores
28 graduating attributes (developing values) of stu- for a student.
dents. Lastly, [56] predicts master’s course remarks,
thus recommending a suitable course to students’ 5) Expert Systems
using singular value decomposition (SVD) and CF. A few studies, comprising 5% of the total papers,
focus on developing expert systems for course ad-
3) Classification / Rule-based / Neural Network vising. For instance, [78] suggests a rule-based ex-
Supervised learning methods such as classification pert system by employing Oracle Policy Automa-
and rule-based schemes are also employed to rec- tion (OPA) software to assist undergraduate stu-
ommend. Seven papers, representing 16% of the dents in academic course selection. [63] proposes
total studies, are based on different classification an educational advisory system by employing fuzzy
and rule-based techniques. [33] implements the K- logic into an expert system, and [65] presents the
nearest neighbor (KNN) algorithm to recommend a Course Advisory Expert System (CAES), consist-
list of graduate schools to the users. Moreover, [38] ing of rule-based reasoning (RBR) and case-based
uses K-nearest Neighbor, Random Forest, SVM reasoning (CBR).
separately on the training data and founds SVM
performs better than the other two. Similarly, [57] 6) Swarm Intelligence
compares the performance of the Linear Regression Among all the studies only one study [64] compares
Model, Naïve Bayes, Support Vector Machine, K- five different swarm intelligence algorithms, e.g.,
Nearest Neighbor, and Decision Tree Classifiers Ant Colony Optimization (ACO), Particle Swarm
to identify appropriate courses goes with students’ Optimization (PSO), Intelligent Weed Optimization
grade. However, [45] involves extracting rules from (IWO), Bee Colony Optimization (BCO), and Bat
the preferences of previous students and thus rec- Algorithm (BA) for student course recommenda-
ommending by checking the similarity between the tion. Among them, ACO applies a hybrid CF and
courses pursued by the students and the precedents content-based filtering approach, whereas others de-
of the rules. Besides, Neural networks are also pop- pend on CF.
ular, and their usage is reported to be promising
for both hybrid and single methods. Specifically, 7) Semantic Analysis and Graph-Based Approach
the proposed framework, intelligent recommenda- Alternatively, [79] involves a series of algorithms,
tion system (IRS) [51] employs multi-layered feed including TF-IDF, word2vec, latent semantic analy-
forward NN to recommend the appropriate courses sis (LSI), and Doc2vec utilized for similarity anal-
where success chance is higher, and [55] recom- ysis based course recommendation. Course Rank
mends courses at the graduate level by applying [80], provides course recommendations based on
feed-forward neural networks. the Ford-Fulkerson max-flow algorithm with con-
straints (e.g., prerequisites, requirements, etc.) fo-
4) Clustering cused on satisfying the degree program require-
A few studies—less than 0.7%—refer to using a ments at each semester. Moreover, [81] proposes
clustering-based approach. [32] designs a college a Personalized Forecasting Model (N-PSEF and
recommender system by utilizing a weighted clus- BPSEF), a personalized recommendation system for
tering process, WCLUSTER. The authors in [41] academic curriculum paths based on performance
propose an ontology-based framework to distin- traits, learning styles, and cognitive traits for each
guish students’ interests and skills in order to pro- individual.
vide a recommendation for a university. However, C. TYPES OF DATA SOURCES USED RS
the proposed framework needs to be evaluated. RQ3 seeks to know the types of input parameters
Moreover, a framework to recommend course en- that are used in developing RS. As seen in Table
rollment based on clustering techniques is explored VII, various input feature groups are used in the
in [44]. Apart from that, [28] applies a K-means design of RS. All the research articles comprised
clustering strategy with different numbers of clus- at least two groups of parameters. Predominantly,
ters using Principal Component Analysis (PCA) 23 studies, representing 41% of the total studies,
to identify students with similar preferences and combine more than one type of data in design-
VOLUME 11, 2023 9
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
ing RS. Academic data comprises university/college entrepreneurial spirit & service orientation, demo-
major, GPA, course grades, and entrance score. De- graphic data, study preferences, goal orientation,
mographic Data such as age, gender, and ethnicity communication style, multilingual proficiency, tech-
are proven to be crucial parameters in these stud- nology adoption, cultural preferences, time man-
ies. Moreover, 12 primary studies use one or more agement, prioritization skills, course data. In Table
demographic parameters. Furthermore, the course 7 and 8 we have shown each of these parameters
recommender system mainly comprises course data, significance and relevance in the establishment of
e.g., name, type, credit, department, description, course recommendation system. In the previous
instructor information, and ratings of each course studies each of these parent parameters was splitted
enrolled or preferences. These data categories rep- into several child parameters. For each of those child
resent 36% and 14%, respectively. parameters a test was conducted. The parent param-
Recommenders for graduate school are mainly eter’s score was the summation of corresponding
taking into account Performance data (standard test child parameters scores.
scores (GRE, TOEFL), research publication, work Let’s denote:
experience) and Institute Profile (location, type, P as the parent parameter’s score,
ranking, safety, facility, admission requirements), Ci as the score of the ith child parameter.
which comprise only 3 studies. In some studies The mathematical equation representing the sum-
(around 13% of all studies), RS are designed based mation of the child parameters’ scores for the parent
on ratings of user interest (personal, academic, or parameter would be:
professional) and several aspects of an institute’s X
profile, such as location, facilities, and faculties. P = Ci (1)
A few studies utilized parameters from more than i
two categories. [43] uses user interest, GPA, and This equation expresses that the parent parameter’s
test scores. [36] combines personality type with score (P) is the sum of the scores of all correspond-
other parameters, and [18] consists of data like ing child parameters (Ci ).
the skills of learners, family income, and institute In Table 9, 10, 11, 12, 13, 14 and 15 we have
profile. In [28], the authors utilize age, gender, represented each of the parent parameter’s corre-
high school GPA, and the score of seven subjects sponding child parameters.
on the university entrance exam, along with the
university elective course description and grade. In D. DEVELOPMENT PLATFORM OF RS
this study, we have observed that, total 24 parent RQ4 seeks information on whether the proposed so-
parameters was used in the course recommendation lutions in primary studies are an algorithm/series of
system. The parent parameters are: self-regulatory algorithms, a web- or mobile-based application, or a
learning strategies, user interest, approach towards framework. The analysis of RQ4 in Table VIII illus-
learning, performance data, psychosocial contextual trates that most reviewed papers propose systems,
factors, institute profile, personality dimensions, which represent 39.3% of the total of the studies,
skills of learners, motivational factors, academic followed by studies corresponding to methods, with
data, cognitive preferences, ratings, learning styles, 38.81% then
10 VOLUME 11, 2023
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
TABLE 7: Significance and relevance of each parameter in the course recommendation System
Parameters Significance Relevance
Self-Regulatory Indicates the student’s ability to regulate their Courses that encourage self-directed learning
Learning own learning processes, including goal-setting, and provide opportunities for reflection may be
Strategies self-monitoring, and self-reflection. recommended.
Approach Reflects the student’s general attitude and style Helps in recommending courses that align
Towards Learning in acquiring knowledge and skills. with the preferred learning style, whether it’s
through hands-on experiences, theoretical anal-
ysis, or creative exploration.
Psychosocial Takes into account the social and psychological Recommends courses that consider the stu-
Contextual aspects influencing learning, including family dent’s overall well-being and align with their
Factors support, peer relationships, and external stres- social and psychological context.
sors.
Ratings Ratings data gauges quality and satisfaction, Ratings data shapes a dynamic system, en-
informing future recommendations in a course hancing precision and relevance for improved
system. learning experiences in recommendations.
Skills of learners Learners’ skills feature is significant for precise Skills of learners contribute to a personalized,
alignment in course recommendations. skill-focused system for an effective educa-
tional journey.
Personality Assesses various personality traits that can im- Courses can be recommended based on how
Dimensions pact academic preferences and performance. well they match the student’s personality, en-
suring a better fit for their individual strengths
and preferences.
Performance data Performance data, including test scores and Performance data shapes a targeted course rec-
work experience, is significant for precise uni- ommendation system based on individual apti-
versity course recommendations. tude and experiences.
Motivational Fac- Identifies what drives the student to learn, Recommends courses that align with the stu-
tors whether it’s intrinsic passion or extrinsic re- dent’s motivations, whether they are driven by
wards. personal interest or future career prospects.
Institute data Institute data tailors course recommendations Institute data shapes a context-aware recom-
based on unique attributes of institutions. mendation system, aligning suggestions with
institution features for informed decision-
making.
Cognitive Prefer- Highlights the student’s preferred thinking and Recommends courses that cater to the student’s
ences problem-solving approaches. cognitive strengths, such as analytical or cre-
ative thinking.
Learning Styles Identifies how the student best absorbs infor- Recommends courses with teaching methods
mation, whether visually, auditorily, or kines- that align with the student’s preferred learning
thetically. style.
Demographic Demographic data tailors academic sugges- Demographic data contributes to a personalized
data tions based on individual backgrounds, enhanc- and inclusive course recommendation system,
ing the system’s adaptability. aligning with unique user needs and socio-
economic contexts.
Entrepreneurial Indicates the student’s inclination towards in- Recommends courses that foster
Spirit & Service- novation, risk-taking, or a desire to make a entrepreneurial skills or align with service-
Orientation social impact. oriented professions.
Academic data Academic data shapes recommendations based Academic data shapes a precise course sys-
on strengths and study preferences in a course tem, aligning with study programs, academic
system. achievements, and preferences.
Study Preferences Reflects whether the student prefers indepen- Recommends courses that match the stu-
dent study or collaborative research. dent’s study preferences, promoting a con-
ducive learning environment.
TABLE 8: Significance and relevance of each parameter in the course recommendation System
Parameters Significance Relevance
Goal Orientation Identifies the student’s short-term and long- Recommends courses that contribute to the
term goals. achievement of the student’s goals, whether
they are immediate objectives or future aspira-
tions.
User interest data User interest data tailors suggestions, fostering User interest data creates a personalized, user-
engagement and satisfaction in a course sys- centric system for an engaging educational ex-
tem. perience.
Communication Assesses the student’s comfort with verbal and Recommends courses that align with the stu-
Style written communication. dent’s communication strengths, be it through
spoken or written expression.
Multilingual Pro- Highlights the student’s language skills and Recommends language-related courses or
ficiency interest in multicultural communication. courses that involve cross-cultural perspectives.
Technology Assesses the student’s comfort with technology Recommends courses that leverage technology
Adoption and adaptability to technological changes. or match the student’s preference for traditional
or online learning.
Course data Course data shapes tailored suggestions align- Course data ensures a comprehensive recom-
ing with academic offerings in a recommenda- mendation system, finely tuned to details and
tion system. contributing to informed decision-making for
students.
Cultural Identifies the student’s cultural sensitivity, Recommends courses that incorporate diverse
Preferences global awareness, and interest in cultural stud- cultural perspectives or align with the student’s
ies. cultural interests.
Time Assesses the student’s ability to manage time Recommends courses that consider the stu-
Management effectively. dent’s time management skills and preferences
for flexible schedules.
Prioritization Reflects the student’s ability to prioritize tasks Recommends courses that align with the stu-
Skills and manage multiple responsibilities. dent’s prioritization skills, ensuring a balanced
workload.
studies comprised frameworks, with 19.7%, and in [29], [33], [49] the authors deployed their system
finally algorithms, with 19% of the total primary into Android apps.
studies. Some of the studies have deployed the
recommendation systems on specific platforms. The
E. EMPIRICAL VALIDATION OF RS
web is the most widely adopted platform for de-
veloping a recommendation system. 11 studies de- RQ5 examines the validation criteria used in these
ployed the RS framework on a web platform, and selected studies, which may be theoretical or exper-
imental. Among the validation categories, 4 distinct
12 VOLUME 11, 2023
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
classes have been identified, and 86% of all the stud- conducted experiments to assess the efficacy of a
ies are validated in some way. Table IX depicts that recommender system.
22 papers, which represent 50% of the total studies,
Parent Parameter Child Parameters Parent Parameter Child Parameters Parent Parameter Child Parameters
Age Study Program Course Level
Gender GPA Course Duration
Ethnicity Academic Scores Prerequisites
in Specific
Subjects
Location Research Course Format
Experience
Family Income Internship/Work Credit Hours
Experience
Educational Academic Instructor
Demographic data Academic data Course data
Background Achievements Experience
First-generation Extracurricular Student
College Status Activities Reviews
Language Study Abroad Textbook/Reading
Proficiency Experience Materials
Accessibility Specialized Skills Lab or Practical
Requirements Component
Parental Professional Course Syllabus
Educational Certifications
Background
TABLE 13: Parameters hierarchy table of demographic, academic and course data factors for course
recommendation system.
Parent Parameter Child Parameters Parent Parameter Child Parameters Parent Parameter Child Parameters
Standardized Test Ranking in Personal Interests
Scores Breakdown Specific Programs
Subject-Specific Facility Academic
Test Scores Breakdown Preferences
Entrance Exam Location Career
Section Scores Attributes Aspirations
Research Teacher Profiles Preferred
Publications Learning Style
Work Experience Program-Specific Extracurricular
Details Requirements Involvement
Performance data Institute data User interest
Internship Details Extracurricular Preferred
Opportunities Teaching Methods
Professional Internship Professional
Certifications Placement Record Development Goals
Project Alumni Success Industry Interests
Contributions Stories
Leadership Roles Collaborative Preferred Work
Research Initiatives Environment
Conference Accreditation Technology
Presentations Information Interests
TABLE 14: Parameters hierarchy table of performance, institute data and user interest factors for course
recommendation system.
execution time and number of clusters. The results square error (MSE) to define the accuracy of the
show that the W-clustering algorithm is scalable to models, whereas [52] measures the performance
the maximum extent. In [36], the authors compare of the rating matrix using domain knowledge with
their proposed HOSVD PCA algorithm with MC- precision. [56] uses a statistical accuracy metric, the
IB CF and Higher-order singular value decomposi- mean absolute error (MAE) between the predicted
tion (HOSVD) algorithms by measuring precision, and the real values, to analyze the error in predic-
recall, F1 metric, and execution time and observe tion. [57] measures the accuracies and RMSE of
that the model performs better among them as well the four different classification models. In [58], the
as handles scalability issues efficiently. [43] evalu- authors examine the model by comparing it with
ates several classification models (with or without related work. In [61], the performance of the model
feature selection) based on accuracy. In [45], the is measured using ROC analysis, and using the pa-
authors compares their course recommendation sys- rameters of the analysis, accuracy (.73), sensitivity
tem, RARE, with the other two course recommen- (.68), and specificity (.88) values are computed. The
dation systems (SCR and AACORN) for evaluation root mean squared error (RMSE) is used to evaluate
and finds that their system performs better under the the models in [81].
cold start problem compared to the two previous
systems. In [59], the authors compared their pro- 2) Academic Case-study
posed model and other recommendation approaches
Another important category is validating a proposed
by average precision and mean reciprocal rank
method using an academic case study, which repre-
(MRR). Moreover, [60] compares the traditional
sents 26.7% of all the studies. [46] uses evaluation
CBF, CF, and proposed OPCR algorithms by three
based on real data from students with respective
performance measure metrics: recovery, accuracy of
letter grades for courses. Also, [26] validates the
relevance, and rank accuracy for online evaluation
system through a case study using MAE and con-
to assess the results obtained from the participants.
fusion matrix analysis. In [48], the authors evaluate
Conversely, [33] evaluates the system by measuring
the performance of their proposed solution using
its accuracy with the variation of training and test
a dataset of academic records of university stu-
data. Similarly, [47] performs descriptive analysis
dents. Using Area under the ROC Curve (AUC)
on the experiment results, while [49] evaluates their
as a performance metric, [50] compares their pro-
mobile-based college recommender with RMSE and
posed BPR-MF solutions with four different sets
accuracy metrics. Studies [51] and [54] utilize mean
of models, such as baseline, memory-based, graph-
16 VOLUME 11, 2023
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
based, and ensemble of different types of solutions, a mean satisfaction level of 3.89 out of 5.0, which
using a real-world course registration dataset. In indicates 77.8% user satisfaction.
[53], the authors use an enrollment dataset from a However, a number of studies [37], [38], [40],
university to compare the accuracy of the proposed [42], [44], [79] did not validate their proposed so-
Markov-based model with item-based and matrix lutions.
factorization-based course recommenders. The eval-
uation of this proposed model in [27] is performed F. OUTCOME OF RS
using a real course dataset of graduate electrical RQ6 aims to discover studies that are not only
engineering students by computing the precision implemented, but also assist students in making
and recall of different variations of the following informed decisions by providing real-life academic
parameters: minimum confidence, minimum match, recommendations. Some studies within the course
minimum specified grade, and minimum support. recommendation domain are successful platforms.
Conversely, Janusz Sobecki [64] compares the per- They are used by several universities to provide
formance of the 5 SI algorithm by utilizing a course- students with personalized and appropriate recom-
grade dataset using the following metrics: Mean mendations. They require planning their academic
Absolute error (MAE), Normalized Mean Absolute path. For example, RARE [45] an association rule-
Error (NMAE), and prediction accuracy (PA). ACO based course recommender system, is used at the
performed best with 0.88 PA. [66] uses the RMSE University de Montréal. Also, CourseRank [80], a
score to evaluate the student performance predic- personalized Stanford University Curriculum, now
tion model and the Simple Matching Coefficient provides course-related services, e.g., choosing the
(SMC) as the study-path recommendation model’s right courses for each student at many universities
efficiency evaluation. In [80], the authors involve throughout the United States. Another one is a per-
558 undergraduate students at Stanford to evaluate sonalized forecasting model, a performance predic-
their package recommendations based on Precision. tion system, and a course recommendation system
More specifically, we can say that the researchers used at CanTho University [81].
around the world are interested in educational rec-
ommendation system, mostly in USA, India, China, V. RESULT SYNTHESIS
Saudi Arabia, and Thailand.
Within the educational domain, recommender sys-
tems serve numerous purposes by generating mean-
3) Survey ingful recommendations from an abundance of in-
Some studies—around 17.6% of all studies—validate formation. In this study, we surveyed the state-of-
the RS through a survey. For instance, [34] and [35] the-art within the domain of mainstream educational
conducted a survey where students compared their choice recommendation systems over the last ten
admitted college results with the proposed system years. Applying systematic review methodology, a
prediction. Also, [46] conducted a survey to deter- total of 56 research papers were identified and ex-
mine the satisfaction of the students upon seeing the amined from an initial set of 1578 studies. This
recommendation to measure the effectiveness of the section presents the empirical findings of this re-
system. Some studies not only used performance view and provides insights based on the overall
metrics to evaluate the systems but also validated analysis of the selected papers. The findings have
them using surveys. The effectiveness of [59] was outlined three distinct vital axes of empirical re-
measured using 1,000 test cases by comparing the search on academic choices: university selection,
Top-3 courses generated by their proposed recom- program selection, and course selection. We have
mender with three other popular courses. Another inspected the frequency of educational choice ar-
study, [60], was also evaluated by a group of uni- ticles and domains from Table VI and found that
versity students. Following the evaluation, students among the disciplines of educational preferences
ranked user satisfaction level and recommendation discussed in this review, course recommendation is
quality. Besides, in [62], the authors "Rule-based a demanding research area, as extensive research
expert systems for supporting university students" is has been carried out in this field [34–59]. How-
tested by Oracle Policy Automation (OPA). Finally, ever, it is noted that research work towards rec-
evaluation of the intelligent advisor (CAES) [65] ommending universities or study programs is min-
is carried out by human advisors, who rate the imal; only 7 studies were found over the last 13
CAES recommendation on a Likert scale of 0–5 to years. Therefore, major/program recommendation
determine the degree of reliability. Results illustrate after high school/college needs a significant focus.
VOLUME 11, 2023 17
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
A. APPLICATION OF HOLLAND CODE (I), Artistic (A), Social (S), Enterprising (E), and
ASSESSMENT IN UNIVERSITY COURSE Conventional (C) categories of the RIASEC model.
RECOMMENDATION SYSTEM This assessment can take the form of an online ques-
After meticulously examining all of the research tionnaire or a traditional paper-based tool, offering
papers explored in this study, we have observed that a comprehensive approach to capturing the diverse
all of these course recommendation system follows facets of a student’s personality. After completion
some non-standard systems relying on individual of the assessment, each student’s responses can
characteristics testing. As a standardized approach, be transformed into a mathematical representation
Holland code assessment can be employed to rec- known as a vector. In this context, a vector is a
ommend courses. multidimensional entity where each dimension cor-
The integration of the Holland Code assessment responds to one of the RIASEC categories. For each
in course recommendation system, offers a com- question, the system captures the student’s response
prehensive solution to the challenge of selecting and updates the corresponding dimension in the
perfect courses. Through a meticulous mathematical vector. For example, the vector representation like
analysis, the system can provide tailored recom- [R, I, A, S, E, C].
mendations, empowering students to make informed Vector normalization for Consistency – Vector nor-
decisions about their academic paths. Applying Hol-
land Code assessments in university course recom-
mendation system involves of assessment outcomes
with the individual student’s academic and career
preferences. The Holland Code, also known as the
RIASEC model, categorizes individuals into six FIGURE 6: Matrix to map RIASEC factor values
personality types: to corresponding courses.
R Realistic
I Investigative malization ensures the vectors maintain consistent
A Artistic scales and do not introduce biases due to variations
S Social in response intensity. The process involves adjust-
E Enterprising ing the values in each vector to a standard range
C Conventional (between 0 and 1). This guarantees that the overall
In below, we have explained the steps how the magnitude of the vector does not influence the sub-
Holland Code assessment could be use in university sequent analyses. The Normalized Value (NV) can
course recommendation system. be calculated using the below equation:
Max Value − Min Value
1) Administer the Assessment NV =
Original Value (OV) − Min Value
In the initial phase of the Holland Code Assess-
ment process, students are presented with a set of 2) Course Mapping
carefully crafted questions intended to discern their Align Courses with Holland Codes – In the course
preferences across the Realistic (R), Investigative mapping phase, the objective is to establish a con-
20 VOLUME 11, 2023
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
TABLE 20: Significance and relevance of Holland Code Assessment’s RIASEC factors in course
recommendation system
Factor name Significance Relevance
Realistic Realistic traits favor hands-on activities, en- In course recommendations, Realistic traits
suring recommendations suit those inclined guide towards programs with tangible
towards applied and skill-based courses. skills, like vocational training or technical
courses.
Investigative Investigative traits emphasize analytics, Investigative traits are pertinent for sug-
guiding recommendations for courses that gesting programs in scientific research, data
cultivate critical thinking and problem- analysis, or disciplines requiring analytical
solving skills. thinking.
Artistic Artistic traits favor creativity, enhancing the Artistic traits guide towards programs nur-
system’s significance by suggesting courses turing creative talents and artistic expres-
in fine arts, design, or creative disciplines. sion.
Social Social traits stress interpersonal Social traits are relevant for suggesting pro-
skills, holding significance in grams in fields like psychology, sociology,
recommending courses involving or any discipline requiring strong interper-
teamwork, communication, and community sonal skills.
engagement.
Enterprising Enterprising traits signify leadership and Enterprising traits are pertinent for suggest-
business acumen, adding significance by ing programs in business, management, or
recommending courses aligned with en- entrepreneurship.
trepreneurial pursuits and leadership devel-
opment.
Conventional Conventional traits stress organization and Conventional traits are relevant for suggest-
detail orientation, providing significance ing programs in fields like finance, admin-
in recommending courses with structured istration, or any discipline requiring organi-
tasks and attention to detail. zational skills.
Moreover, considering the ever-evolving educa- in the material, they are more likely to actively par-
tional landscape, continuous research and innova- ticipate in class discussions, complete assignments,
tion are essential to ensuring that recommender and seek out additional learning opportunities.
systems remain relevant, trustworthy, and support-
ive tools in empowering students to make well- C. RETENTION IMPROVEMENT
informed choices for their educational and profes- Course recommendation systems can contribute to
sional success. higher student retention rates by guiding students
towards courses that match their academic strengths
VII. PRACTICAL USES OF COURSE and interests. When students feel supported in their
RECOMMENDATION SYSTEM academic journey and see a clear path towards their
Course recommendation systems are essential tools goals, they are less likely to drop out.
for individuals navigating the complex landscape
of career path selection. These systems provide D. REDUCED DROPOUT RATES
personalized guidance by leveraging advanced data Decreasing dropout rates is a key benefit of
analytics and machine learning algorithms to ana- course recommendation systems. By helping stu-
lyze individual preferences, skills, and career aspi- dents choose courses that align with their abilities
rations. By offering tailored recommendations for and interests, these systems can prevent academic
courses and educational pathways, they assist users overwhelm and increase the likelihood of successful
in exploring diverse career options and identifying course completion.
the most suitable educational opportunities. Further-
more, course recommendation systems help bridge
E. COURSE DIVERSITY PROMOTION
the gap between individuals’ current skills and the
requirements of various industries by suggesting Promoting diversity in course selection is essential
courses that align with emerging trends and work- for providing a well-rounded education. Course rec-
force demands. This proactive approach ensures that ommendation systems can achieve this by suggest-
users acquire the necessary expertise to succeed in ing a variety of options from different disciplines,
their chosen fields and remain competitive in the job cultures, and perspectives, encouraging students to
market. Additionally, these systems facilitate life- explore new subjects and broaden their horizons.
long learning by encouraging users to continuously
update their skills and knowledge, enabling them to F. ADDRESSING CURRICULUM GAPS
adapt to evolving job requirements and pursue ca- Course recommendation systems can identify gaps
reer advancements effectively. Overall, course rec- in the curriculum and suggest supplementary or
ommendation systems play a vital role in empow- complementary courses to fill those gaps. This en-
ering individuals to make informed decisions about sures that students receive a comprehensive educa-
their career paths, facilitating professional growth, tion that covers all necessary topics and prepares
and enhancing overall career satisfaction and suc- them for future academic or professional endeavors.
cess. Here, we have discussed various practical uses
of course recommendation system in personalized G. OPTIMIZING COURSE LOAD
career path selection process. Helping students optimize their course load is cru-
cial for academic success. Course recommendation
A. PERSONALIZED LEARNING PATHS systems can suggest a balanced mix of core, elec-
Tailoring course recommendations based on indi- tive, and prerequisite courses, taking into account
vidual student preferences and academic goals en- students’ schedules, academic goals, and extracur-
ables students to pursue a curriculum aligned with ricular commitments.
their interests and career aspirations. This fosters a
sense of ownership over their education, increasing H. ADAPTIVE LEARNING SUPPORT
motivation and engagement. Adaptive learning environments require personal-
ized instruction tailored to each student’s needs.
B. IMPROVED STUDENT ENGAGEMENT Course recommendation systems play a key role
By suggesting courses that align with students’ in supporting adaptive learning by recommending
interests, learning styles, and career goals, course courses that adapt to students’ evolving knowledge
recommendation systems can enhance student en- and skill levels, ensuring they receive appropriate
gagement. When students are genuinely interested challenges and support.
VOLUME 11, 2023 23
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
. ALIGNMENT WITH INDUSTRY NEEDS faculty can identify areas for improvement and share
Course recommendation systems can align course best practices with colleagues, fostering a culture of
offerings with industry needs and workforce de- continuous improvement and innovation.
mands. By analyzing labor market trends and em-
ployer feedback, institutions can identify areas of . ALIGNMENT WITH INSTITUTIONAL MISSION
growth and opportunity and tailor course recom- AND VALUES
mendations accordingly. Course recommendation systems should align with
the institutional mission and values to ensure that
. SUPPORT FOR ACADEMIC PLANNING recommended courses reflect the institution’s ed-
COMMITTEES ucational philosophy and goals. By incorporating
Academic planning committees rely on course rec- institutional priorities and priorities into course rec-
ommendation data to make decisions about curricu- ommendations, these systems contribute to the over-
lum development, program evaluation, and resource all coherence and integrity of the academic program.
allocation. By providing timely and accurate course
recommendations, these systems support the work . SUPPORT FOR TRANSFER ARTICULATION
of academic planning committees and help institu- AGREEMENTS
tions achieve their strategic goals. Course recommendation systems can support trans-
fer articulation agreements by recommending
. INTEGRATION WITH STUDENT SUPPORT courses that meet transfer requirements and articu-
SERVICES lation guidelines. By facilitating the transfer process
Course recommendation systems can integrate with for students, these systems promote seamless path-
student support services, such as tutoring, advising, ways to degree completion and academic success.
and counseling, to provide holistic support to stu-
dents. By identifying students who may benefit from . COMMUNITY ENGAGEMENT AND OUTREACH
additional assistance and connecting them with the Course recommendation systems can engage the
appropriate resources, these systems contribute to broader community, including alumni, employers,
student success and well-being. and industry partners, in the educational process. By
soliciting input and feedback from external stake-
. INTERNATIONALIZATION AND GLOBAL holders, institutions can ensure that course offerings
ENGAGEMENT are relevant, responsive, and aligned with commu-
Course recommendation systems can support in- nity needs and priorities.
ternationalization efforts by recommending courses These are some practical uses highlighting the mul-
that reflect diverse perspectives and global trends. tifaceted impact of course recommendation sys-
By exposing students to a variety of cultural and tems on student success, institutional effectiveness,
linguistic experiences, these systems prepare them and educational innovation. By leveraging data-
to thrive in an increasingly interconnected world. driven insights and personalized recommendations,
these systems support informed decision-making,
. ALIGNMENT WITH ACCREDITATION enhance learning experiences, and promote equi-
STANDARDS table access to educational opportunities for all stu-
Course recommendation systems can help insti- dents.
tutions demonstrate compliance with accreditation
standards and quality assurance measures. By en- VIII. CONCLUSION
suring that recommended courses meet established In recent years, the field of tertiary education has
criteria for rigor, relevance, and effectiveness, these witnessed a notable surge in interest in develop-
systems support accreditation processes and institu- ing recommender systems for making informed de-
tional accountability. cisions. This systematic review aimed to investi-
gate the trends and techniques employed in rec-
. FACULTY COLLABORATION AND ommendation systems within the context of higher
PROFESSIONAL DEVELOPMENT educational academic advising. By analyzing 56
Course recommendation systems can facilitate fac- selected studies, we gained valuable insights into
ulty collaboration and professional development by the utilization of recommender systems in six key
providing insights into teaching and learning prac- aspects driven by research questions, including the
tices. By analyzing course recommendation data, purpose, development approach, incorporated fea-
26 VOLUME 11, 2023
Nabila et al.: Recommender System in Academic Choices of Higher Education: A Systematic Review
tures, deployment, validation criteria, and educa- dation: a review of ontology- based recommender systems for
tional outcomes. The findings revealed that course e-learning, Artificial intelligence review 50 (2018) 21–48.
[7] S. Garcia-Martinez, A. Hamou-Lhadj, Educational recom-
recommendations emerged as the most prominent mender systems: A pedagogical- focused perspective, Multi-
area of focus, accounting for approximately 53% of media services in intelligent environments: Recommendation
the papers. Notably, hybrid strategies constituted the services (2013) 113–124.
[8] K. Verbert, N. Manouselis, X. Ochoa, M. Wolpers, H. Drach-
primary development technique across the analyzed
sler, I. Bosnic, E. Duval, Context- aware recommender systems
studies. However, it was observed that there is no for learning: a survey and future challenges, IEEE transactions
one-size-fits-all generic model or framework for on learning technologies 5 (4) (2012) 318–335.
recommending educational choices, as each recom- [9] S. Khusro, Z. Ali, I. Ullah, Recommender systems: issues,
challenges, and research op- portunities, in: Information sci-
mender system is tailored to its specific context and ence and applications (ICISA) 2016, Springer, 2016, pp.
data type. As a conclusion, this systematic review 1179–1189.
sheds light on the growing importance of recom- [10] L. Iaquinta, M. De Gemmis, P. Lops, G. Semeraro, M. Fi-
lannino, P. Molino, Introducing serendipity in a content-based
mender systems in academic advising in higher recommender system, in: 2008 eighth international conference
education. The identified trends and techniques can on hybrid intelligent systems, IEEE, 2008, pp. 168–173.
guide future researchers in developing innovative [11] S. C. Cazella, P. A. Behar, D. Schneider, K. K. da Silva, R.
Freitas, Developing a learning objects recommender system
approaches to unlock the full potential of academic
based on competences to education: Experience report, in:
advising and enhance student learning experiences. New Perspectives in Information Systems and Technologies,
By effectively leveraging recommender systems, Volume 2, Springer, 2014, pp. 217–226.
educational institutions can offer personalized guid- [12] M. Erdt, A. Fernandez, C. Rensing, Evaluating recommender
systems for technology en- hanced learning: a quantitative
ance to students, leading to more informed and survey, IEEE Transactions on Learning Technologies 8 (4)
successful academic decisions. The comprehensive (2015) 326–344.
understanding gained from this review can pave [13] J. Zhong, H. Xie, F. L. Wang, The research trends in recom-
mender systems for e-learning: A systematic review of ssci
the way for the continuous improvement and im- journal articles from 2014 to 2018, Asian Association of Open
plementation of recommender systems in tertiary Universities Journal 14 (1) (2019) 12–27.
education, contributing to the advancement of the [14] N. Manouselis, H. Drachsler, R. Vuorikari, H. Hummel, R.
Koper, Recommender systems in technology enhanced learn-
field and ultimately benefiting students’ academic
ing, Recommender systems handbook (2011) 387–415.
journeys. [15] S. Dietze, H. Drachsler, D. Giordano, A survey on linked data
and the social web as facilitators for tel recommender systems,
Recommender Systems for Technology Enhanced Learning:
ACKNOWLEDGMENT Research Trends and Applications (2014) 47–75.
This work was supported by the ICT Division, Min- [16] F. L. da Silva, B. K. Slodkowski, K. K. A. da Silva, S.
istry of Posts, Telecommunications, and Informa- C. Cazella, A systematic literature review on educational
recommender systems for teaching and learning: research
tion Technology, Bangladesh and Institutite of Ad- trends, limitations and opportunities, Education and Informa-
vanced Research (IAR), United International Uni- tion Technologies 28 (3) (2023) 3289–3328.
versity under grant number IAR-2024-Pub-012. We [17] K.-K. Chu, M. Chang, Y.-T. Hsia, Designing a course recom-
mendation system on web based on the students’ course selec-
express gratitude to our colleagues from the Ad- tion records, in: EdMedia+ Innovate Learning, Association for
vanced Intelligent Multidisciplinary Systems Lab the Advancement of Computing in Education (AACE), 2003,
(AIMS Lab), United International University (UIU), pp. 14–21.
Bangladesh, whose insights and expertise greatly [18] V. Vaidhehi, R. Suchithra, A systematic review of recom-
mender systems in education, Inter- national Journal of En-
enriched this study. gineering & Technology 7 (1) (2018).
[19] S. A. A. Maria, S. C. Cazella, P. A. Behar, Sistemas de
recomendac ao: conceitos e t ecnicas de aplicac ao, Recomen-
REFERENCES dac ao Pedag ogica em Educac ao a Dist^ancia (2019) 19–47.
[1] R. Rafeh, Recommender systems in ecommerce, 2017. [20] H. Drachsler, K. Verbert, O. C. Santos, N. Manouselis,
[2] G. Linden, B. Smith, J. York, Amazon. com recommendations: Panorama of recommender systems to support learning, Rec-
Item-to-item collaborative filtering, IEEE Internet computing 7 ommender systems handbook (2015) 421–451.
(1) (2003) 76–80. [21] A. C. Rivera, M. Tapia-Leon, S. Lujan-Mora, Recommenda-
[3] P. Covington, J. Adams, E. Sargin, Deep neural networks for tion systems in education: A systematic mapping study, in:
youtube recommendations, in: Proceedings of the 10th ACM Proceedings of the International Conference on Information
conference on recommender systems, 2016, pp. 191–198. Technology & Systems (ICITS 2018), Springer, 2018, pp.
[4] C. Gomez-Uribe, N. Hunt, The netflix recommender system: 937–947.
Algorithms, business value, and innovation, in “acm transac- [22] E. Ashraf, S. Manickam, S. Karuppayah, A comprehensive
tion on management information systems”, vol. 6 (2016). review of course recommender systems in e-learning., Journal
[5] M.-I. Dascalu, C.-N. Bodea, M. N. Mihailescu, E. A. Tanase, P. of Educators Online 18 (1) (2021).
Ordo nez de Pablos, Ed- ucational recommender systems and [23] C. Salazar, J. Aguilar, J. Monsalve-Pulido, E. Montoya, Af-
their application in lifelong learning, Behaviour & information fective recommender systems in the educational field. a sys-
technology 35 (4) (2016) 290–297. tematic literature review, Computer Science Review 40 (2021)
[6] J. K. Tarus, Z. Niu, G. Mustafa, Knowledge-based recommen- 100377.
[24] B. Kitchenham, O. P. Brereton, D. Budgen, M. Turner, J. study, International Journal of e-Education, e-Business, e-
Bailey, S. Linkman, Systematic literature reviews in software Management and e-Learning 4 (5) (2014) 329.
engineering–a systematic literature review, Information and [42] C. Obeid, I. Lahoud, H. El Khoury, P.-A. Champin, Ontology-
software technology 51 (1) (2009) 7–15. based recommender system in higher education, in: Compan-
[25] P. Brereton, B. A. Kitchenham, D. Budgen, M. Turner, M. ion proceedings of the the web conference 2018, 2018, pp.
Khalil, Lessons from applying the systematic literature review 1031–1034.
process within the software engineering domain, Journal of [43] M. Vukicevic, M. Jovanovic, B. Delibasic, M. Suknovic, Rec-
systems and software 80 (4) (2007) 571–583. ommender system for selection of the right study program
[26] P.-C. Chang, C.-H. Lin, M.-H. Chen, A hybrid course rec- for higher education students, RapidMiner: Data Mining Use
ommendation system by integrating collaborative filtering and Cases and Business Analytics Applications (2013) 145.
artificial immune systems, Algorithms 9 (3) (2016) 47. [44] Q. E. Booker, A student program recommendation system
[27] A. Al-Badarenah, J. Alsakran, An automated recommender prototype, Issues in Information Systems (2009) 544–551.
system for course selection, International Journal of Advanced [45] N. Bendakir, E. A ımeur, Using association rules for course
Computer Science and Applications 7 (3) (2016) 166– 175. recommendation, in: Proceedings of the AAAI workshop on
[28] S. Asadi, S. Jafari, Z. Shokrollahi, Developing a course recom- educational data mining, Vol. 3, Citeseer, 2006, pp. 1–10.
mender by combining clustering and fuzzy association rules, [46] A. Slim, D. Hush, T. Ojha, C. Abdallah, G. Heileman, G. El-
Journal of AI and Data mining 7 (2) (2019) 249–262. Howayek, An automated framework to recommend a suitable
[29] K. Pupara, W. Nuankaew, P. Nuankaew, An institution recom- academic program, course and instructor, in: 2019 IEEE Fifth
mender system based on student context and educational insti- International Conference on Big Data Computing Service and
tution in a mobile environment, in: 2016 International Com- Applications (BigDataService), IEEE, 2019, pp. 145–150.
puter Science and Engineering Conference (ICSEC), IEEE, [47] K. Ganeshan, X. Li, An intelligent student advising system
2016, pp. 1–6. using collaborative filtering, in: 2015 IEEE Frontiers in Ed-
[30] A. Baskota, Y.-K. Ng, A graduate school recommendation ucation Conference (FIE), IEEE, 2015, pp. 1–8.
system using the multi-class support vector machine and knn
[48] K. Bhumichitr, S. Channarukul, N. Saejiem, R. Jiamthap-
approaches, in: 2018 IEEE International Conference on Infor-
thaksin, K. Nongpong, Recom- mender systems for university
mation Reuse and Integration (IRI), IEEE, 2018, pp. 277–284.
elective course recommendation, in: 2017 14th international
[31] S. Fong, R. P. Biuk-Aghai, An automated university admission joint conference on computer science and software engineering
recommender system for secondary school students, in: The (JCSSE), IEEE, 2017, pp. 1–5.
6th international conference on information technology and
[49] Z. Khoja, S. Shetty, Hybrid recommender system for college
applications, 2009, p. 42.
courses, in: 2017 International Conference on Computational
[32] M. Y. S. Reddy, P. Govindarajulu, College recommender sys-
Science and Computational Intelligence (CSCI), IEEE, 2017,
tem using stu- dent’preferences/voting: A system development
pp. 1167–1171.
with empirical study, Int. J. Comput. Sci. Netw. Secur 18
(2018) 87–98. [50] E. L. Lee, T.-T. Kuo, S.-D. Lin, A collaborative filtering-based
two stage model with item dependency for course recommen-
[33] M. Hasan, S. Ahmed, D. M. Abdullah, M. S. Rahman, Gradu-
dation, in: 2017 IEEE international conference on data science
ate school recommender system: Assisting admission seekers
and advanced analytics (DSAA), IEEE, 2017, pp. 496–503.
to apply for graduate studies in appropriate graduate schools,
in: 2016 5th International Conference on Informatics, Elec- [51] K. Kongsakun, C. C. Fung, Neural network modeling for an
tronics and Vision (ICIEV), IEEE, 2016, pp. 502–507. intelligent recommendation system supporting srm for univer-
[34] A. H. M. Ragab, A. F. S. Mashat, A. M. Khedra, Hrspca: sities in thailand, WSEAS transactions on Computers 11 (2)
Hybrid recommender system for predicting college admission, (2012) 34–44.
in: 2012 12th International conference on intelligent systems [52] V. Vaidhehi, R. Suchithra, An enhanced approach using col-
design and applications (ISDA), IEEE, 2012, pp. 107–113. laborative filtering for generating under graduate program
[35] A. H. M. Ragab, A. F. S. Mashat, A. M. Khedra, Design recommendations, in: 2019 Second International Conference
and implementation of a hybrid recommender system for pre- on Advanced Computational and Communication Paradigms
dicting college admission, International Journal of Computer (ICACCP), IEEE, 2019, pp. 1–6.
Information Systems and Industrial Management Applications [53] E. S. Khorasani, Z. Zhenge, J. Champaign, A markov chain
6 (2014) 35–44. collaborative filtering model for course enrollment recommen-
[36] D. K. Bokde, S. Girase, D. Mukhopadhyay, An approach to a dations, in: 2016 IEEE International Conference on Big Data
university recommendation by multi-criteria collaborative fil- (Big Data), IEEE, 2016, pp. 3484–3490.
tering and dimensionality reduction techniques, in: 2015 IEEE [54] B. Bakhshinategh, G. Spanakis, O. Zaiane, S. ElAtia, A course
International Symposium on Nanoelectronic and Information recommender system based on graduating attributes, in: Inter-
Systems, IEEE, 2015, pp. 231–236. national Conference on Computer Supported Education, Vol.
[37] Z. Khanam, S. Alkhaldi, An intelligent recommendation en- 2, SCITEPRESS, 2017, pp. 347–354.
gine for selecting the university for graduate courses in ksa: [55] J. Britto, S. Prabhu, A. Gawali, Y. Jadhav, A machine learn-
Sars student admission recommender system, in: Inventive ing based approach for recom- mending courses at graduate
Computation Technologies 4, Springer, 2020, pp. 711–722. level, in: 2019 International Conference on Smart Systems and
[38] J. Manley, S. Krishnakumar, University recommender system Inventive Technology (ICSSIT), IEEE, 2019, pp. 117–121.
for graduate studies in usa. [56] F. O. G. Carballo, Masters’ courses recommendation: Explor-
[39] S. Fong, Y.-W. Si, R. P. Biuk-Aghai, Applying a hybrid model ing collaborative filtering and singular value decomposition
of neural network and decision tree classifier for predicting with student profiling, no. November (2014).
university admission, in: 2009 7th international conference on [57] M. Isma’il, U. Haruna, G. Aliyu, I. Abdulmumin, S. Adamu,
information, communications and signal processing (ICICS), An autonomous courses recom- mender system for undergrad-
IEEE, 2009, pp. 1–5. uate using machine learning techniques, in: 2020 international
[40] G. Meryem, K. Douzi, S. Chantit, Toward an e-orientation conference in mathematics, computer engineering and com-
platform: Using hybrid recom- mendation systems, in: 2016 puter science (ICMCECS), IEEE, 2020, pp. 1–6.
11th International Conference on Intelligent Systems: Theories [58] A. Esteban, A. Zafra, C. Romero, Helping university students
and Applications (SITA), IEEE, 2016, pp. 1–6. to choose elective courses by using a hybrid multi-criteria rec-
[41] K. Wakil, B. Akram, N. Kamal, A. Safi, Web recommender ommendation system with genetic optimization, Knowledge-
system for private universities’ admission in iraq: Uhd case Based Systems 194 (2020) 105385.
[59] Y.-K. Ng, J. Linn, Crsrecs: a personalized course recommen- [76] Z. Wu, Q. Liang, Z. Zhan, Course recommendation
dation system for college stu- dents, in: 2017 8th International based on enhancement of meta-path embedding in
Conference on Information, Intelligence, Systems & Appli- heterogeneous graph, Applied Sciences (2023). URL
cations (IISA), IEEE, 2017, pp. 1–6. https://api.semanticscholar.org/CorpusID:256880819
[60] M. E. Ibrahim, Y. Yang, D. L. Ndzi, G. Yang, M. Al-Maliki, [77] M. Gao, Y. Luo, X. Hu, Online course rec-
Ontology-based personalized course recommendation frame- ommendation using deep convolutional neural
work, IEEE Access 7 (2018) 5180–5199. network with negative sequence mining, Wireless
[61] M. F. Adak, N. Yumusak, H. Taskin, An elective course sug- Communications and Mobile Computing (2022). URL
gestion system developed in computer engineering department https://api.semanticscholar.org/CorpusID:251421701
using fuzzy logic, in: 2016 International Conference on Indus- [78] M. Rajabi, S. Hossani, F. Dehghani, A literature review on
trial Informatics and Computer Systems (CIICS), IEEE, 2016, current approaches and applica- tions of fuzzy expert systems,
pp. 1–5. arXiv preprint arXiv:1909.08794 (2019).
[62] G. Engin, B. Aksoyer, M. Avdagic, D. Bozanlı, U. Hanay, D. [79] H. Ma, X. Wang, J. Hou, Y. Lu, Course recommendation
Maden, G. Ertek, Rule-based expert systems for supporting based on semantic similarity analysis, in: 2017 3rd IEEE
university students, Procedia Computer Science 31 (2014) International Conference on Control Science and Systems
22–31. Engineering (ICCSSE), IEEE, 2017, pp. 638–641.
[63] M. H. Goodarzi, V. Rafe, Educational advisor system imple- [80] A. Parameswaran, P. Venetis, H. Garcia-Molina, Recommen-
mented by web-based fuzzy expert systems (2012). dation systems with complex constraints: A course recommen-
[64] J. Sobecki, Comparison of selected swarm intelligence al- dation perspective, ACM Transactions on Information Sys-
gorithms in student courses rec- ommendation application, tems (TOIS) 29 (4) (2011) 1–33.
International Journal of Software Engineering and Knowledge [81] N. Thai-Nghe, T. Horv, L. Schmidt-Thieme, et al., Personal-
Engineering 24 (01) (2014) 91–109. ized forecasting student perfor- mance, in: 2011 IEEE 11th In-
[65] O. Daramola, O. Emebo, I. Afolabi, C. Ayo, Implementation ternational Conference on Advanced Learning Technologies,
of an intelligent course advisory expert system, International IEEE, 2011, pp. 412–414.
Journal of Advanced Research in Artificial Intelligence 3 (5) [82] M. S. Sulaiman, A. S. A. Tamizi, M. R. Shamsudin, A. Azmi,
(2014) 6–12. Course recommendation system using fuzzy logic approach,
[66] T. Vuong, T. Trinh, H. Ha, X. Phan, T. Tran, A personalized Indonesian Journal of Electrical Engineering and Computer
course recommender system for undergraduate students, Inter- Science 17 (2020) 365–371.
national Journal of Learning and Teaching ijlt (2019) 181–190. [83] N. Torres, Recommender systems for education: A case of
[67] M. R. Sundari, G. Shreya, T. Jawahar, Course recommendation study using formative assessments, 2022 41st International
system, International Journal of Computer Applications 175 Conference of the Chilean Computer Science Society (SCCC)
(2020) 13–16. (2022) 1–6.
[68] F. L. Roux, E. Ranjeet, V. Ghai, Y. Gao, J. Lu, A [84] V. Sankhe, J. Shah, T. Paranjape, R. Shankarmani, Skill based
course recommender system using multiple criteria course recommendation system, 2020 IEEE International Con-
decision making method, in: Proceedings of the 2007 ference on Computing, Power and Communication Technolo-
International Conference on Intelligent Systems and gies (GUCON) (2020) 573–576.
Knowledge Engineering (ISKE 2007), Atlantis Press,
[85] S. P. Perumal, K. Arputharaj, G. Sannasi, Fuzzy family tree
2007/10, pp. 1407–1411. doi:10.2991/iske.2007.238. URL
similarity based effective e- learning recommender system,
https://doi.org/10.2991/iske.2007.238
2016 Eighth International Conference on Advanced Comput-
[69] H. Bydzovsk a, Course enrollment recommender system, in:
ing (ICoAC) (2017) 146–150.
Educational Data Mining, 2016.
[86] S. Ghosh, A hybrid programming course recommenda-
[70] V. Anupama, M. S. Elayidom, Course recommendation sys-
tion system using fuzzy logic and xdeepfm, 2023 Interna-
tem: Collaborative filtering, machine learning and topic mod-
tional Conference on Intelligent Systems, Advanced Com-
elling, 2022 8th International Conference on Advanced Com-
puting and Communication (ISACC) (2023) 1–8. URL
puting and Communication Systems (ICACCS) 1 (2022)
https://api.semanticscholar.org/CorpusID:257935912
1459–1462.
[87] Q. Guo, F. Zhuang, C. Qin, H. Zhu, X. Xie, H. Xiong, Q. He,
[71] N. D. Lynn, A. W. R. Emanuel, A review on recommender
A survey on knowledge graph-based recommender systems,
systems for course selection in higher education, IOP Con-
IEEE Transactions on Knowledge and Data Engineering 34
ference Series: Materials Science and Engineering 1098 (3)
(2020) 3549–3568.
(2021) 032039. doi:10.1088/1757-899X/1098/3/032039. URL
https://dx.doi.org/10.1088/1757-899X/1098/3/032039 [88] V. G. M. Murillo, D. E. P. Avenda no, F. R. Lopez, J. M.
[72] K. K. Jena, S. K. Bhoi, T. K. Malik, K. S. Sahoo, N. Z. G. Calleros, A systematic literature review on the hybrid ap-
Jhanjhi, S. Bhatia, F. Amsaad, E- learning course recom- proaches for recommender systems, Computaci on y Sistemas
mender system using collaborative filtering models, Elec- (2022).
tronics 12 (1) (2023). doi:10.3390/electronics12010157. URL [89] W. Kaiss, K. Mansouri, F. Poirier, Personalized e-learning
https://www.mdpi.com/2079-9292/12/1/157 recommender system based on a hybrid approach, 2022 IEEE
[73] G. George, A. M. Lal, A personalized approach to Global Engineering Education Conference (EDUCON) (2022)
course recommendation in higher educa- tion, Int. 1621–1627.
J. Semantic Web Inf. Syst. 17 (2021) 100–114. URL [90] G. Deepak, I. Trivedi, A hybridized deep learning
https://api.semanticscholar.org/CorpusID:233988383 strategy for course recommendation, International
[74] Y. Shen, H. Li, Z. Liao, Online education course recom- Journal of Adult Education and Technology (2023). URL
mendation algorithm based on path factors, 2022 IEEE 5th https://api.semanticscholar.org/CorpusID:258282835
International Conference on Information Systems and Com- [91] S. Sharma, A. T. Sharma, Y. Sharma, M. K. Bhatia, Rec-
puter Aided Education (ICISCAE) (2022) 257–260. URL ommender system using hybrid approach, 2016 International
https://api.semanticscholar.org/CorpusID:253271353 Conference on Computing, Communication and Automation
[75] P. Mishra, V. Jain, Course recommendation (ICCCA) (2016) 219–223.
system using content-based filtering, 2023 7th [92] O. Mazhoud, A. Kalboussi, A. H. Kacem, Educational recom-
International Conference on Trends in Electronics mender system based on learner’s annotative activity, Int. J.
and Informatics (ICOEI) (2023) 1431– 1436. URL Emerg. Technol. Learn. 16 (2021).
https://api.semanticscholar.org/CorpusID:258869504 [93] Z. Gulzar, A. A. Leema, G. Deepak, Pcrs: Personalized course
recommender system based on hybrid approach, Procedia DR. FARHANA SARKER received
Computer Science 125 (2018) 518–524. her PhD degree in Computer Science
[94] M. Tkalcic, L. Chen, Personality and recommender systems, from the University of Southampton,
in: Recommender systems handbook, Springer, 2015, pp. UK. Before her PhD study, she worked
715–739. as a researcher in the University of
[95] M. Zhang, J. Tang, X. Zhang, X. Xue, Addressing cold start Southampton, UK and in King Saud
in recommender systems: A semi-supervised co-training algo- University, Riyadh, Saudi Arabia. Dr.
rithm, SIGIR ’14, Association for Computing Machinery, New
Sarker has been an Assistant Professor
York, NY, USA, 2014, p. 73–82.
at Department of Computer Science
[96] I. Palomares, S. V. Kovalchuk, Multi-view data approaches in
recommender systems: an overview: (invited paper), Procedia and Engineering, University of Liberal
Computer Science 119 (2017) 30–41, 6th International Arts Bangladesh (ULAB), Bangladesh from 2015 to 2023. She
Young Scientist Conference on Computational Science, is also a co-founder of CMED Health Ltd, a health tech startup
YSC 2017, 01-03 November 2017, Kotka, Finland. in Bangladesh. She has diversified research background in the
doi:https://doi.org/10.1016/j.procs.2017.11.157. URL area of computer science and engineering. Her current research
https://www.sciencedirect.com/science/article/pii/S187705091 includes education, technology-enhanced teaching and learning,
7323670 linked data, health, and data science. She published a number of
[97] L. Xue, P. Zhang, A. Zeng, Enhancing the long-term per- peer-reviewed journal and conference papers.
formance of recommender system, ArXiv abs/1904.00672
(2019).
[98] M. Fang, J. Liu, M. Momma, Y. Sun, Fairroad: Achieving
fairness for recommender systems with optimized antidote
data, Proceedings of the 27th ACM on Symposium on Access
Control Models and Technologies (2022).
[99] B. P. Knijnenburg, M. C. Willemsen, Z. Gantner, H. Soncu, C.
Newell, Explaining the user experience of recommender sys-
tems, User Modeling and User-Adapted Interaction 22 (2012) ARIFUR RAHMAN received the
441–504. B.Sc. degree in computer science and
[100] B. Cheng, Y. Zhang, D. Shi, Ontology-based personal- engineering from the Khulna Uni-
ized learning path recommendation for course learning, 2018 versity of Engineering and Technol-
9th International Conference on Information Technology in ogy (KUET), Bangladesh, in 2023.
Medicine and Education (ITME) (2018) 531–535. He is currently engaged as an NLP
Research Engineer serving at AIMS
LAB, United International University
with support from the Bill & Melinda
Gates Foundation and ICT Division,
Bangladesh. Besides, he was lead author in several manuscripts
in various conferences. His research interests include deep
learning, bioinformatics, natural language processing, computer
vision and the Internet of Things (IoT).