4.

1 Types of recommender systems

In the following paragraph, we will describe the most popular types of recommender systems in the field of e-
learning

4.1.1 Content-based recommender systems

Content-based recommender systems (CBRS) recommends to the user items similar to those favored in the past
[1]. CBRS is dependent on comparing the learner profile with the LO [2]. In 2015, [3] proposed architecture of a
content-based recommender system as represented in Fig 2. This architecture contains three components:
Content analyzer, Profile learner, and Filtering component [3]. Later [4] updated this architecture by adding a
fourth component named Profile cleaner, the novel architecture is shown in Fig 3.

Fig.2 High-level architecture of content-based recommender [3].

Fig 3 Architecture of a semantics-aware content-based recommender system [4].

4.1.2 Collaborative Filtering Recommender Systems

Collaborative Filtering Recommender Systems (CFRS) recommends to the active user the items that other users
with similar tastes previously preferred [5], in other words, CFRS aids people to make choices based on the
evaluations of other people [6]. The taste similarity of two users is calculated based on the similarity in the
users' rating history [7]. There are many techniques to implement CFRS, however, the most popular ways are
memory-based collaborative filtering and model-based collaborative filtering [8–16], while some authors like
add a third type named hybrid-based collaborative filtering. The difference between CFRS and CBRS is shown
in Fig 4.
 Fig.4 The difference between CFRS, and CBRS [17].

4.1.3 Hybrid-based recommender systems

Hybrid-based recommender systems HBRS is a combination of content-based, and collaborative filtering ,

[5,11,14,17–25] which aids in averting some limitations of collaborative filtering and content-based
[1,21,25,26].
According to [9,17,21,23,26–35], there are seven approaches of hybrid-based recommender systems, which are:
Weighted approach, Switching approach, Feature combination approach, Feature augmentation approach,
Mixed approach, and Meta-level approach.

4.2 Comparison of recommender system types

Content-Based RS Collaborative filtering RS Hybrid-Based RS

Advantages ● The user’s data are ● Capability to recommend ● Combines all the
not required to items outside their advantages of
recommend items preferences to the user content-based and
[23,35,36]. collaborative filtering
and may like this item.
● A new item can be [17].
recommended based [17]. ● Overcomes the cold
on the similarity ● The domain knowledge start, data sparsity,
between items’ of the user doesn’t and
specifications require making overspecialization
[17,35]. predictions [23]. problems [17,36].

Disadvantages  Overspecialization ● Cold start [35–37]. ● Implementation is

problem [26,36]. ● Data sparsity [35–37]. very expensive [36].
● Necessity to analyze ● Scalability [35–37]. ● Hybridization
and discover all item
increases the
characteristics to
create a complexity of the
 recommendation list recommendation
[17]. process [36].

4.4 Recommender systems research challenges

The recommender systems have several challenges, and limitations (Abbas et al., 2015; George & Lal, 2019;
Melville & Vikas, 2010; Mohamed et al., 2019; Patel & Patel, 2015; Sharma et al., 2017; Sharma & Anju, 2013;
Sharma & Gera, 2013; Sinha & Dhanalakshmi, 2019; Su & Khoshgoftaar, 2009; Taneja & Anuja, 2018; Thorat
et al., 2015; Zuva et al., 2012). In the following paragraph, we will present some of them, and in Table 2 we will
introduce certain solutions proposed to overcome these issues.

4.4.1 Cold start

One of the most well-known problems in recommendation systems is the cold start problem (Jeevamol &
Renumol, 2021; Lika et al., 2014). The cold start issue concerns the situation where a new user/element enters
the system (Sharma & Anju, 2013; Zuva et al., 2012). It is divided into two categories: Item cold start, and User
cold start (Jeevamol & Renumol, 2021; Melville & Vikas, 2010; Sharma et al., 2017; Sharma & Anju, 2013;
Sinha & Dhanalakshmi, 2019; Taneja & Anuja, 2018; Zuva et al., 2012). It is really very hard to supply
recommendation because in the case of new user, there is very less information about user that is available and
as well for a new item, no ratings are usually available (Sharma & Gera, 2013; Zuva et al., 2012).

4.4.2 Sparsity

The sparsity issue is one of the main problems that recommender systems encounter and data sparsity have a
major impact on the quality of the recommendation (Sharma & Gera, 2013). It ensues because the user interacts
with a small part of items in the particular application area (Natarajan et al., 2020).

4.4.3 Scalability

With the tremendous growing of information over the internet, it is evident that the recommender systems have
a data explosion, and therefore it is a big challenge to cope with tremendously continually growth demand
(Sharma & Gera, 2013). The traditional recommender systems algorithms suffer from serious scalability issues
as computational resources exceed practical or acceptable levels (Su & Khoshgoftaar, 2009; Zuva et al., 2012).
Some of them deal with the computations that increase as the number of users and items (Sharma & Gera,
2013).

4.4.4 Over Specialization

Users are limited to receive recommendations as defined in their profiles in certain cases and this is known as
the over specialization issue. It prevents the user from discovering new items and other available options
(Mohamed et al., 2019; Sharma & Gera, 2013).

Table 2: Solutions proposed

Authors Challenge Solution

Stitini et Over-specialization Developing a new approach to address the problem of over-specialization

al., 2022. in content-based recommender systems and to create new elements for
the user.

Alhijawi et Cold start and data Developing new model named inheritance-based prediction (INH-BP)
al., 2021. sparsity. that surmounts issues and challenges using the user interest print (UIP)
matrix.

Alhijawi, Cold start Developing OptNibor which is a recommender system that reduces the
2019. impact of the cold start issue and enhances the quality of CFRS.

Mahadeva Data sparsity Proposing an upgraded collaborative filtering algorithm to overcome data
n & sparsity issue, in the videos' domain. The algorithm applies credible user-
Michael, reviews to generate precise recommendations
2017.

Ahmadian Cold start and data Proposing an approach that mitigates the cold start and data sparsity
et al., sparsity issues in recommendation systems by adding trusty virtual ratings.
2018.

Elkahky et Scalability Proposing a solution to the scalability issue by using deep neural
al., 2015. networks to get low dimensional characteristics from high dimensional
ones

Adamopou Over-specialization Presenting a novel probabilistic method for recommending items, that
los & reduces the common issues of over-specialization and concentration bias
Alexander, of recommendations.
2014.

Gemulla et Scalability Suggesting a distributed algorithm where most computations can be

al., 2011. performed on several machines in parallel.

Shanshan Cold start and data Proposing hybrid recommended approach that could aid relieve the cold
et al. 2021. sparsity problems. start and data sparsity issues.

Tarus et Cold start and data Developing hybrid approach can reduce both the cold start and data
al., 2017. sparsity problems. sparsity problems by using ontological domain knowledge and learners'
sequential access pattern before the initial data to work on is available in
the recommender system.