Implementing a recommender system for CS undergraduate students using machine learning
Abstract
E-learning is a learning process that accesses educational curriculum using electronic
technologies and can be operated anywhere in the world where there is facility to
connect to the internet. Traditional teaching practices of face-to-face mentoring
are being replaced by the non-concrete classroom where instructors and students
can interact without any barriers. Prerecorded videos, eBooks (Electronic books in
pdf form), short written lessons, live sessions, video calling are the main sources
of E-learning. Lecturers can grade students' performance through virtual assign-
ments and tests. Students may even opt for a degree certi cate after completion
of the course that are no less worthy than a degree from any renowned physical
institution. In this research paper, the study of E-learning is divided in two parts.
Firstly, a survey was conducted on undergraduate students enrolled in Department
of Computer Science and Engineering, BRAC University. After performing statisti-
cal data mining on students' reviews, more useful information were interpreted that
include the preferred online source to study, satisfaction extent on current obtain-
able resources, any suggestions that could make their E-learning process e ortless
and complaints against current online accessibility of course materials. Whether
the students want an E-learning Recommender System was also deduced after this
assessment. Secondly, an E-learning Recommender System containing video tuto-
rials was built using content-based ltering, item-based collaborative ltering and
user-based collaborative ltering. This recommender system was built using tools
and libraries of Python programming language which contains massive resources for
major CSE courses o ered in BRAC University. The system has also attempted to
eliminate the problems attained from the initial portion of the research. Ultimately,
we proposed a hybrid ltering approach for our video recommender system consid-
ering our experimental results carried on the particular demography which revealed
accuracy of the three used algorithms as 88%, 80% and 80%.