IEEE Systems, Man and Cybernetics Magazine - April 2023 - 8

characteristics used in the literature. The column " Data
Source " shows how researchers collect data related to key
indicators; " logs " denotes that data are from online educational
systems. Table 1 is analyzed as follows:
◆ Most of the literature forms heterogeneous learning
teams, and the knowledge level is often used as a key
indicator.
◆ Most of the literature obtains key indicators via a quiz.
◆ Classic algorithms, e.g., k-means and genetic algorithms,
are often utilized for solving LTF problems.
LL-LTF Methods
LL-LTF methods usually form learning teams based on
learning goals, learners' cognitive ability, learning style,
and so on. Chen et al. [4] claim that all teams should have
balanced learning characteristics, and the heterogeneity of
learners' knowledge state and learning roles could be
guaranteed. Based on that, they present a heterogeneous
LL-LTF approach via a genetic algorithm. A penalty function
is adopted to ensure that members have similar learning
characteristics and social interactions. Moreno et al.
[20] use a genetic algorithm-based approach to form heterogeneous
teams by considering learners' knowledge
state, communication competence, and leadership skill.
Sánchez et al. [21] present a homogeneous LTF approach
via a genetic algorithm. They use the Big Five model [22] to
model learners' personalities and use them as the evaluation
metrics for team formation.
In addition, Noorani et al. [5] propose a mechanism and
an instructional design to foster well-organized collaborative
learning via game theory. Since game theory allows
learners to see their teammates' efforts, they may choose
Table 1. A summary of related work.
Type
of LTF
LL-LTF
Representative
Literature
[4]
[20]
[21]
[5]
LC-LTF
[6]
[24]
[25]
[7]
[8]
[9]
Team Type
Heterogeneous
Heterogeneous
Homogeneous
Heterogeneous/
homogeneous
Heterogeneous
Heterogeneous
Heterogeneous
Heterogeneous/
homogeneous/hybrid
Homogeneous
Homogeneous/heterogeneous
8
Key
Indicators
Knowledge level, team roles, and
social interaction
Knowledge level, communicative
skills, and leadership
Personality
Personal willingness
Competency and willingness
Knowledge level
Knowledge level
Personal characteristics, learning
behaviors, and context information
Learning path
Learning perspectives and
personality
IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE April 2023
Data
Source
Logs
Quiz
Quiz
Quiz
Quiz
Quiz
Solving Techniques
Genetic algorithm
Genetic algorithm
Genetic algorithm
Self-selection based on game
theory
Crowding evolutionary algorithm
k-means and greedy algorithm
Quiz Novel k-means and greedy algorithm
Quiz
Clustering algorithm based on
particle swarm optimization
Logs
Quiz
k-means algorithm
Novel k-means algorithm
to escape the free rider problem by changing their teammates
and persuading them to make more effort to complete
the activities.
LC-LTF Methods
LC-LTF methods first select n outstanding learners as
leaders for n teams and then assign other learners to the
teams according to the principle of complementation and
balance. Yannibelli et al. [6] use Belbin's model to analyze
learners' matches in nine roles and then apply a crowded
evolutionary algorithm to form heterogeneous learning
teams. Quick learners could gain, as teaching others and
giving help have been shown to be positively correlated to
an increase in ability [23]. Based on this, given a class having
many students, each exhibiting a different ability level,
Agrawal et al. [24] present a framework for studying LTF
from a computational perspective, propose a formal definition
of the LTF problem, and investigate some of its variants.
Liu et al. [25] model learners' ability via cognitive
diagnosis analysis, and learners with good knowledge
states are selected as team leaders.
Clustering algorithms are also utilized to deal with LCLTF.
A clustering algorithm classifies learners into several
clusters, i.e., learning teams, and each cluster has a center.
The center of each cluster might be a real learner or a virtual
leader. Zervoudakis et al. [7] randomly select learners
as centers, and all learners are clustered based on 14 characteristics
related to personal characteristics, learning
behaviors, and context information. By applying the
k-means algorithm, Ramos et al. [8] develop a framework
to form homogeneous teams through their learning paths.
Kanika et al. [9] cluster learners from similar learning
IEEE Systems, Man and Cybernetics Magazine - April 2023

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - April 2023
