Identifying Gaps in Use of and Research on Adaptive Learning
Systems
Shuai Wang
1,* a
, Claire Christensen
1,*
, Elizabeth McBride
1,* b
, Hannah Kelly
1
, Wei Cui
2
,
Richard Tong
2
, Linda Shear
1
, Louise Yarnall
1
and Mingyu Feng
3c
1
SRI Education, SRI International, 1100 Wilson Blvd., Suite 2800, Arlington, VA 22209, U.S.A.
2
Squirrel AI Learning by Yixue Education Group, 39 Hongcao Road, Shanghai, China
3
WestEd, 730 Harrison Street, San Francisco, California 94107, U.S.A.
Keywords: Adaptive Learning, Algorithms, Artificial Intelligence, Efficacy Studies, Literature Review.
Abstract: Adaptive learning systems have become increasingly common across age groups and content areas. Many
adaptive learning systems personalize the learning experience based on students’ prior knowledge,
preferences, learner profile, system usage, learning style, and/or learning perceptions. In addition, various
learning algorithms have been developed over the years, such as item response theories, Markov modelling,
recurrent neural networks, and Bayesian knowledge tracing (BKT). Although western countries have
generated numerous efficacy studies, Chinese adaptive education is in its earliest stage, with few efficacy
studies conducted in this context, which is a gap in this field. This position paper describes one Chinese
adaptive learning system, Squirrel AI Learning, and invites further research on adaptive learning systems in
China and other Asian countries.
1 INTRODUCTION
Adaptive learning systems have become increasingly
more popular across age groups and content areas. As
adaptive learning system use has grown, so too has
research on their efficacy for learning. This paper
details gaps in the research and in the use of adaptive
learning systems. After discussing adaptive learning
systems in general, we address a case study on
Squirrel AI Learning in greater depth.
Adaptive learning systems use comprehensive
data analytics and machine-learning algorithms to
provide individualized, computer-based learning
experiences. Many adaptive learning systems
personalize the learning experience based on
students’ prior knowledge, preferences, learner
profile, system usage, learning style, and/or learning
perceptions (Nakic et al., 2015; Xie et al., 2019). As
students spend more time in the system, the system
may develop a more detailed understanding of
a
https://orcid.org/0000-0002-4983-9558
b
https://orcid.org/0000-0002-0168-5705
c
https://orcid.org/0000-0001-9635-1611
*
Sam Wang, Claire Christensen, and Elizabeth McBride contributed to the study equally.
students’ needs and preferences, resulting in greater
personalization (Hauger & Köck, 2007; Van Seters et
al., 2012). Aspects of the system that may be
personalized vary, but often include the learning
sequence, item difficulty, and learning supports
provided.
For many years, computer scientists and cognitive
scientists have developed adaptive learning systems
that use artificial intelligence to mimic the
interactions of one-on-one human tutoring (Merrill,
Reiser, Ranney, & Trafton, 1992). Developers have
created systems that present content, pose questions,
assign tasks, provide hints, answer questions, and
suggest improvements to learners based on their prior
behaviors (Ma, Adesope, Nesbit, & Liu, 2014).
Adaptive learning systems follow a similar “closed
loop” architecture that gathers data from the learner
and then uses that data to estimate the learner’s
progress, recommend activities, gives hints, or
provides tailored feedback. The adaptive system’s
algorithms typically make such decisions by referring
118
Wang, S., Christensen, C., McBride, E., Kelly, H., Cui, W., Tong, R., Shear, L., Yarnall, L. and Feng, M.
Identifying Gaps in Use of and Research on Adaptive Learning Systems.
DOI: 10.5220/0009590701180124
In Proceedings of the 12th International Conference on Computer Supported Education (CSEDU 2020) - Volume 1, pages 118-124
ISBN: 978-989-758-417-6
Copyright
c
 2020 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

to a domain model of the knowledge to be learned, a
student model of learners’ background characteristics
(knowledge level, affect, and motivation), and a task
model that specifies features of the learning activities
(e.g., questions, tasks, quizzes, dynamic hints,
feedback, prompts, and recommendations) (Lee &
Park, 2008).
Adaptive learning systems utilize various
algorithms, such as item response theories, Markov
modelling, recurrent neural networks, Bayesian
knowledge tracing (BKT), natural language
processing, and other machine learning models to
personalize the learning sequence for each student.
This personalization is based on system-generated
student profiles, informed by a students’ performance
on an initial knowledge diagnostic and continuously
updated with student usage data and learning
behaviors. As students spend more time in the system,
their learning profiles become more accurate and
allow for greater personalization (Hauger & Köck,
2007; Van Seters et al., 2012).
Historically, adaptive learning systems utilize
different models for estimating a learner’s level of
domain knowledge. Some focus on how well learners
implement core steps in tasks (Koedinger, Anderson,
Hadley, & Mark, 1997); some examine learners’
problem-solving behaviors and compare them to
models of both correct and incorrect behaviors
(Aleven, McLaren, Sewall, & Koedinger, 2009);
others present specific problem-solving situations
and check for common errors (Mitrovic, 2012;
Ohlsson, 1992). Some use natural language
processing to measure how well learners articulate
common learning goals or misconceptions (Graesser,
VanLehn, Rose, Jordan, Harter, 2001), while still
others model a learners’ understanding at each point
of interaction (Piech et al, 2015; Yudelson,
Koedinger & Gordon, 2013).
Often, the content area being assessed is linked to
the requirements for the algorithm that is used. For
example, in content areas where knowledge
components can be clearly distinguished, as in
algebra, a model like Bayesian knowledge tracing
may work well for modelling student behaviour and
providing guidance for learning. However, in a
content area, like English, where writing a coherent
argument is important, natural language processing is
an important algorithm to include. Use of knowledge
components is also a common metric used to trace
student understanding in adaptive learning systems.
A knowledge component is a description of a mental
structure a student uses to accomplish steps in a
problem or task (Koedinger, Corbett, & Perfetti,
2012). A knowledge component relates features of a
question or task to a response given by a student.
2 EFFICACY RESEARCH
Efficacy studies from the past 6 years have shown
that the use of adaptive learning systems is associated
with greater gains in student learning. Across several
different types of studies, including a review of 37
adaptive learning efficacy studies, a comparative
study of 1,600 adaptive and 4,800 non-adaptive
courses, and a large-scale randomized control trial in
algebra classrooms, researchers report positive
findings when they compare students who use
adaptive learning systems in academic contexts to
those who do not. In two studies that involved the
implementation of adaptive learning systems in
mathematics instruction, students’ scores on
proficiency exams were 8 and 3 percentile points
higher, respectively, than the scores of their peers
who did not use the learning system (Bomash & Kish,
2015; Pane et al., 2014, 2017).
Despite these promising findings, we know
relatively little about for whom and in what contexts
adaptive learning systems are most effective. Below
we provide an overview of the relevant literature to
date and highlight areas for future research and
development.
2.1 Learner Characteristics
Adaptive learning systems have potential to promote
equity in education by responding to diverse learners’
needs in ways that may not be feasible for teachers in
traditional classroom instruction. We call for more
research on subgroup effects to evaluate whether and
how adaptive learning realizes this promise. Do
adaptive learning systems work equally well for all
learners? Or are they better suited for a particular
level of prior knowledge, socioeconomic status,
gender, or age range?
Researchers have begun to explore the differential
impacts of prior knowledge on students’ learning
from adaptive learning systems. Many adaptive
learning systems personalize the learning experience
based on prior knowledge (Nakic et al., 2015), as
students with different prior knowledge may benefit
from different instructional features (Ayres, 2006;
Flores et al., 2012). For example, to avoid cognitive
overload in learners with less prior knowledge,
adaptive learning systems may tailor worked
examples or the degree of learner autonomy (Lee et
al., 2008; McNeill et al., 2006; Salden et al., 2009;
Identifying Gaps in Use of and Research on Adaptive Learning Systems
119

Scheiter et al., 2007). Some evidence suggests that
such adaptations can be effective: some studies have
found that adaptive learning is able to help students
with less prior knowledge achieve similar outcomes
to students with more prior knowledge (Jones, 2018;
Wang et al., 2019). More research is needed to
demonstrate which adaptive learning algorithms and
features are best suited to addressing the needs of
learners with varying prior knowledge.
In addition to adapting to differences in prior
knowledge, adaptive learning may also have potential
to address equity gaps that affect historically
underserved populations, such as students from lower
socioeconomic-status (SES) families. These students
tend to report lower academic participation and
academic achievement on math and reading
assessments (Dahl & Lochner, 2012; Willms, 2003).
Initial research suggests that such students may
benefit from adaptive learning systems in academic
settings. A study by Yarnall and colleagues (2016)
assessed 2-year and 4-year college students’
impressions of adaptive learning systems in higher
education and found that 2-year college students, who
are more likely to be lower SES, rate adaptive
learning systems more favourably. More of these
students also reported positive learning gains than
their 4-year institution peers. In two cases reported in
this study, Pell grant students showed similar positive
learning outcomes associated with using adaptive
courseware to those reported for the general
population. More research is needed to explore
adaptive learning systems’ promise with lower SES
students, and to elucidate the most effective
adaptations for this population, such as those that
address prior knowledge gaps and learner motivation.
More research is needed on the efficacy of
adaptive learning systems for K–12 students. Most
adaptive learning efficacy studies are conducted with
higher education students. A systematic review of 70
articles on adaptive and personalized learning
published from 2007 to 2017 found that the largest
share, 46%, included higher education students. Less
than half that, 21% of studies, included elementary
students, and only 9% included middle and high
school students (Xie et al., 2019). While K–12
efficacy studies in adaptive learning are less common,
some suggest adaptive learning can be effective for
elementary students (Mettler et al., 2011) and middle
school students (Feng et al., 2018).
2.2 Content Areas and Features
A systematic review of 70 articles on personalized
and adaptive learning found that many studies are
conducted in certain content areas, while in others
there have been few studies. Xie (et al, 2019) report
that the most studies are conducted in engineering and
computer science, followed by languages,
mathematics, and science. Content areas like health
(medical/nursing), social science, art/design, and
business management have few or no efficacy studies
using adaptive learning systems. While there is a
larger proportion of research conducted on adaptive
learning systems in more traditional school content
areas (e.g., math and science), there is little research
done on adaptive learning systems in content areas
that are more commonly found in college courses or
professional training (e.g., health/medical/nursing).
This is consistent with prior studies (e.g. Alexander,
Rose, & Woodhead, 1992) that discussed the limited
domain knowledge of researchers who developed
adaptive learning products.
In addition, Xie (et al, 2019) report on the types of
learning supports provided by adaptive learning
systems. Personalized learning content is the most
common feature, followed by personalized learning
paths, personalized interfaces, personalized diagnosis
and suggestions, personalized recommendations, and
personalized prompts or feedback. While many of
these personalization features have been studied in
some way, a meta-analysis of their impact on student
learning has not been conducted. Since adaptive
learning systems rarely contain only one way of using
personalization, a meta-analysis of these features is
particularly necessary as it will provide insight into
design features for these systems. However, due to
the large range of learner populations, content areas,
and system features contained under the adaptive
learning systems umbrella, many more studies on
efficacy must be conducted.
2.3 Outcomes
As mentioned previously, adaptive learning has been
shown to have significant positive impacts on
learning outcomes. More research is needed to
explore how effects vary by outcome. For instance,
consistent with prior studies (Brookhart, 2020;
White, 1993), Xie (et al, 2019) have found that
positive effects of adaptive learning are more likely
to be reported on student affect than on student
cognition More research is needed to distinguish
which learning metrics are most sensitive to adaptive
learning-related change. While some studies measure
academic outcomes solely via in-system performance
(Bomash & Kish, 2015; Jones, 2018), others have
examined adaptive learning’s effects on external
metrics.
CSEDU 2020 - 12th International Conference on Computer Supported Education
120

Test scores and assessments are some of the most
common external metrics that researchers use to
assess the efficacy of adaptive learning systems.
Studies that measure student outcomes on
assessments have shown mixed, though mostly
positive, effects (Fullerton & Hughes, 2016; Yarnall
et al., 2016). A review of six studies found that the
use of adaptive learning systems in various
environments (upper elementary through the
workplace) related to positive effects on students’
course test achievement (Durlach & Ray, 2011). An
important feature of the adaptive learning systems in
these studies that may have contributed to the positive
results is mastery-based progression, which requires
students to demonstrate their knowledge before
advancing. Another study review by Yarnall and
colleagues (2016) reported that while it positively
impacted students’ scores on course tests, the use of
adaptive courseware did not have significant effects
on course grades or completion rates.
3 GLOBAL SPREAD
Adaptive learning systems are increasingly common
in western culture, including the United States,
United Kingdom, and other regions. Some well-
known products already on the market for learners
and educators include Knewton (Wilson & Nichols,
2015), ASSISTments (Heffernan & Heffernan,
2014), ALEKS (Canfield, 2001), i-Ready,
Achieve3000, Carnegie Learning, Norton, Kidaptive,
and DreamBox Learning. Accordingly, many
efficacy studies of adaptive learning systems have
been only conducted in western countries (e.g., Griff
& Matter, 2013; Mettler et al., 2011; Sun et al., 2017).
Meanwhile, in mainland China, adaptive learning
systems are only just beginning to gain popularity,
even though 19% of all Chinese Internet users have
engaged in online education in recent years (China
Internet Network Information Center, 2017). While
adaptive learning is relatively new in mainland China,
it is a Chinese education policy priority (O’Meara,
2019).
3.1 Squirrel AI Learning
Squirrel AI Learning is considered the first
commercial adaptive learning system in mainland
China. Since its establishment in 2016, Squirrel AI
Learning has expanded to serve almost 2 million
registered accounts in over 700 cities. These 2 million
users are diverse in socioeconomic status, urbanicity,
and academic achievement. In addition to this user
base, Squirrel AI Learning has opened over 2,000
learning centers in less than 5 years. This rapid
expansion is indicative of the gap that adaptive
learning fills in the Chinese after-school tutoring
market.
In particular, during the coronavirus outbreak in
2020, Squirrel AI Learning is considered an
important supplement for student learning while
students are required to stay at home and schools are
closed in almost all provinces in China. Because of
their potential to reach so many students, research is
needed to ensure that such learning systems truly
support student learning in China and possibly in
other countries of Asia.
3.2 Squirrel AI Learning Features
Squirrel AI Learning provides instructions and
supports for K–12 students and has the following
features:
1. Nanoscale Knowledge Components.
Squirrel AI Learning breaks down
knowledge components into a fine-grained
knowledge map with knowledge
components organized hierarchically based
on the following: relationship to learning
progression; adaptive diagnostic pre-
assessment; automated differentiated
instruction; rich, high-quality learning
repository of various types of learning
content; immediate feedback and
explanations to students and in-class support
and intervention by teachers. For example,
in junior high school mathematics, 300
knowledge components are dissolved into
30,000 fine-grained knowledge
components, and each knowledge
component is matched with the learning
content. This content may include text items,
animation, slides, short instructional videos,
etc. A parent knowledge component can be
resolved into sub-knowledge components
that are more specific and targeted. See
figure 1 for an example of the Squirrel AI
Learning system, where students receive
popup explanations if their attempts are
incorrect; students can choose to view video
or text explanations.
2. Integration of Various Learning
Algorithms. Squirrel AI Learning uses
more than 10 learning algorithm
technologies, including Clustering
algorithm, such as k-means and expectation
maximization (EM), logistic regression,
Identifying Gaps in Use of and Research on Adaptive Learning Systems
121

Item Response Theory, graph theory,
probabilistic graph model, Bayesian
network, knowledge space theory,
information theory, source tracing model,
knowledge tracking theory, learning
analysis technology, and so on. The
algorithms help to identify students'
weaknesses in current knowledge by tracing
the pre-requisite knowledge components of
their current learning content. The
algorithms determine the recommendation
priority of each knowledge component from
three aspects: whether the pre-requisite
knowledge component is easy to learn,
whether its map position is relatively
backward, and its central degree. This
layered process of identifying student
weaknesses creates a solid foundation to
effectively promote the learning of current
and future knowledge components.
3. MCM Model (Methodology, Capacity,
and Mode of Thinking). Squirrel AI
Learning splits MCM into nanoscale just
like the knowledge components, and the
ambiguous and incomprehensible
capabilities are split into nanoscale
capabilities that can be clearly defined. In
this way, Squirrel AI Learning can measure
the level of a student's capabilities and
represent his capabilities quantitatively. At
the same time, Squirrel AI Learning ensures
that the capabilities can not only be clearly
explained by teachers but also understood
and digested by students. MCM are curated
and summarized; for example, in middle
school mathematics, 500 components
subdivide into 1,000 application scenarios to
make them completely definable,
measurable and teachable. Squirrel AI
Learning can identify and quantify the
capabilities that individuals possess and the
capabilities that they need to improve, such
as those of a lawyer who is excellent in
pattern exploration skills and summarization
skills yet lacks the skills needed for
analyzing 3D graphics, algorithm
construction, and realization. In contrast to
the lawyer, a scientist may be excellent at
data analysis but poor in linguistic
association skills. In Squirrel AI Learning,
the users’ behavior is taken automatically
into account by the algorithm parameters.
However, in the initial phase, some of the
parameters are set according to experts’
experience. For example, the item difficulty
is set according to experts’ experience if the
number of data samples is less than 25.
When the number of data samples is more
than 25, the item difficulty is computed by
machine learning algorithms.
Figure 1: Screenshot of Squirrel AI Learning system.
Note. The question is on the upper left of the white
window. The
choices are on the right side of the
white window. When students answer the question
incorrectly, the system gives explanations at the
bottom of the white window. The next question is
dependent on the correctness and difficulty level of
the current question.
3.3 Squirrel AI Learning Efficacy
Multiple efficacy studies have shown that the Squirrel
AI Learning platform:
 Improves learning efficiency and students’
perceptions of the learning experience, as
compared to other learning platforms (Li,
Cui, Xu, Zhu, & Feng, 2018)
 Is associated with greater improvements in
test scores compared with whole-classroom
instruction or small-group tutoring (Feng et
al. 2018; Wang, Christensen, Cui, Tong,
Yarnall, Shear, & Feng, submitted). This
builds upon a review finding that contrary to
popular belief, intelligent tutoring is nearly
as effective as human tutoring (VanLehn,
2011).
 Is associated with similar gain scores
regardless of students’ prior knowledge
(Wang, Feng, Bienkowski, Christensen, &
Cui, 2019).
However, the sample sizes for these studies were
not large, the topics covered in these studies were
limited, and the interventions of these programs were
short-term, e.g. in a couple of days. Future studies
CSEDU 2020 - 12th International Conference on Computer Supported Education
122

examining Chinese adaptive learning efficacies in
different contexts are highly needed.
4 CONCLUSIONS
Adaptive learning systems have become more widely
used in the last 2 decades and are only becoming more
widely used with each passing year. The ubiquity of
adaptive learning systems demands wide-reaching
studies on their efficacy. Many current efficacy
studies apply adaptive learning systems in higher
education and in traditional academic subjects (math,
science, languages). Further efforts are still needed to
determine which outcome metrics are both useful and
aligned with the use of adaptive learning systems.
Further, more research is needed to address how these
adaptive learning systems might address issues of
equity or otherwise impact lower- SES students.
One additional gap in the research is in study
geography. Many efficacy studies using adaptive
learning systems take place in either the United States
or the United Kingdom. With the potential to impact
many students worldwide, efficacy studies must be
undertaken in a wider variety of contexts. We
discussed one such case, Squirrel AI Learning, a
commercial adaptive learning system in mainland
China. However, efficacy studies on Squirrel AI
Learning have included limited numbers of
participants and a limited range of subject areas.
Efficacy studies of different contexts need to be
conducted. More importantly, as new technological
and pedagogical approaches continue to evolve, more
efficacy studies are needed in Asia and worldwide in
the future, and we invite more scholars to continue
research in the adaptive learning systems field.
REFERENCES
Alexander, R., Rose, J., & Woodhead, C. (1992).
Curriculum organisation and classroom practice in
primary schools: A discussion paper. The Department
of Education and Science of United Kingdom.
Ayres, P. (2006). Impact of reducing intrinsic cognitive
load on learning in a mathematical domain. Applied
Cognitive Psychology, 20(3), 287–298.
https://doi.org/10.1002/acp.1245
Bomash, I., & Kish, C. (2015). The improvement index:
Evaluating academic gains in college students using
adaptive lessons. New York: Knewton.
Brookhart, S. M. (2010). How to assess higher-order
thinking skills in your classroom. ASCD.
Canfield, W. (2001). ALEKS: A Web-based intelligent
tutoring system. Mathematics and Computer
Education, 35(2), 152.
China Internet Network Information Center. (2017). The
40th China Statistical Report on Internet Development.
http://www.cnnic.cn/hlwfzyj/hlwxzbg/hlwtjbg/201708
/P02017080735192326213.pdf
Dahl, G. B., & Lochner, L. (2012). The impact of family
income on child achievement: Evidence from the
earned income tax credit. American Economic Review,
102(5), 1927–56.
Durlach, P. J., & Ray, J. M. (2011). Designing adaptive
instructional environments: Insights from empirical
evidence. Orlando, FL: Army Research Institute for the
Behavioral and Social Sciences.
Feng, M., Cui, W., & Wang, S. (2018). Adaptive learning
goes to China. International Conference on Artificial
Intelligence in Education, 89–93.
Flores, R., Ari, F., Inan, F. A., & Arslan-Ari, I. (2012). The
impact of adapting content for students with individual
differences. Journal of Educational Technology &
Society, 15(3), 251–261.
Fullerton, J., & Hughes, R. (2016). DreamBox learning
achievement growth. Retrieved from Center for
Education Research Policy, Harvard University.
Retrieved from https://cepr.harvard.edu/dreambox-
learning-achievement-growth
Graesser, A. C., VanLehn, K., Rosé, C. P., Jordan, P. W.,
& Harter, D. (2001). Intelligent tutoring systems with
conversational dialogue. AI Magazine, 22(4), 39-51.
Griff, E. R., & Matter, S. F. (2013). Evaluation of an
adaptive online learning system. British Journal of
Educational Technology, 44(1), 170–176.
Hauger, D., & Köck, M. (2007). State of the art of
adaptivity in e-learning platforms. LWA, 355–360.
Heffernan, N. T., & Heffernan, C. L. (2014). The
ASSISTments ecosystem: Building a platform that
brings scientists and teachers together for minimally
invasive research on human learning and
teaching. International Journal of Artificial
Intelligence in Education, 24(4), 470-497.
Jones, A. (2018, March 15). Interpreting Knewton’s 2017
student mastery results.
https://www.knewton.com/blog/mastery/interpreting-
knewtons-2017-student-mastery-results/
Koedinger, K. R., Corbett, A. T., & Perfetti, C. (2012). The
KnowledgeLearningInstruction framework:
Bridging the sciencepractice chasm to enhance robust
student learning. Cognitive science, 36(5), 757-798.
Lee, C.-H., Lee, G.-G., & Leu, Y. (2008). Analysis on the
adaptive scaffolding learning path and the learning
performance of e-learning. WSEAS Transactions on
information science and Applications, 5(4), 11.
McNeill, K. L., Lizotte, D. J., Krajcik, J., & Marx, R. W.
(2006). Supporting students’ construction of scientific
explanations by fading scaffolds in instructional
materials. The Journal of the Learning Sciences, 15(2),
153–191. JSTOR.
Merrill, D. C., Reiser, B. J., Ranney, M., & Trafton, J. G.
(1992). Effective tutoring techniques: A comparison of
Identifying Gaps in Use of and Research on Adaptive Learning Systems
123

human tutors and intelligent tutoring systems. Journal
of the Learning Sciences, 2(3), 277–305.
Mettler, E., Massey, C. M., & Kellman, P. J. (2011).
Improving Adaptive Learning Technology through the
Use of Response Times. Proceedings of the 33
rd
Annual
Meeting of the Cognitive Science Society, 2532-2537.
https://files.eric.ed.gov/fulltext/ED549436.pdf
Mitrovic, A. (2012). Fifteen years of constraint-based
tutors: What we have achieved and where we are going.
User Modeling and User-Adapted Interaction, 22(1-2),
39–72.
Nakic, J., Granic, A., & Glavinic, V. (2015). Anatomy of
student models in adaptive learning systems: A
systematic literature review of individual differences
from 2001 to 2013. Journal of Educational Computing
Research, 51(4), 459–489.
O’Meara, S. (2019). Will China lead the world in AI by
2030? Nature, 572, 427-428 (2019)
Ohlsson, S. (1992). Information-processing explanations of
insight and related phenomena. In M. T. Keane & K. J.
Gilhooly (Eds.), Advances in the psychology of thinking
(pp. 1–44). New York, NY: Harvester Wheatsheaf.
Pane, J. F., Griffin, B. A., McCaffrey, D. F., & Karam, R.
(2014). Effectiveness of cognitive tutor algebra I at
scale. Educational Evaluation and Policy Analysis,
36(2), 127–144.
Pane, J. F., Steiner, E. D., Baird, M. D., Hamilton, L. S., &
Pane, J. D. (2017). How does personalized learning
affect student achievement? Arlington, VA: RAND
Corporation.
Piech, C., Bassen, J., Huang, J., Ganguli, S., Sahami, M.,
Guibas, L. J., & Sohl-Dickstein, J. (2015). Deep
knowledge tracing. In Advances in neural information
processing systems (pp. 505-513).
Salden, R. J. C. M., Aleven, V. A. W. M. M., Renkl, A., &
Schwonke, R. (2009). Worked examples and tutored
problem solving: Redundant or synergistic forms of
support? Topics in Cognitive Science, 1(1), 203–213.
https://doi.org/10.1111/j.1756-8765.2008.01011.x
Scheiter, K., & Gerjets, P. (2007). Learner control in
hypermedia environments. Educational Psychology
Review, 19(3), 285–307.
Sun, Q., Abdourazakou, Y., & Norman, T. J. (2017).
LearnSmart, adaptive teaching, and student learning
effectiveness: An empirical investigation. Journal of
Education for Business, 92(1), 36–43.
KURT VanLEHN (2011): The Relative Effectiveness of
Human Tutoring, Intelligent Tutoring Systems, and
Other Tutoring Systems, Educational Psychologist,
46(4), 197-221.
Van Seters, J. R., Ossevoort, M. A., Tramper, J., &
Goedhart, M. J. (2012). The influence of student
characteristics on the use of adaptive e-learning
material. Computers & Education, 58(3), 942–952.
Wang, S., Feng, M., Bienkowski, M., Christensen, C., &
Cui, W. (2019). Learning from an adaptive learning
system: Student profiling among middle school
students. Proceedings of the 11th International
Conference on Computer Supported Education -
Volume 1: CSEDU, 78–84.
https://doi.org/10.5220/0007729700780084
White, E. M. (1993). Assessing higher-order thinking and
communication skills in college graduates through
writing. The Journal of General Education, 105–122.
Willms, J. D. (2003). Student engagement at school. A
Sense of Belonging and Participation. Paris:
Organisation for Economic Co-Operation and
Development.
Wilson, K., & Nichols, Z. (2015). The Knewton Platform.
A General-Purpose Adaptive Learning
Infrastructure. A Knewton White Paper.
Xie, H., Chu, H.-C., Hwang, G.-J., & Wang, C.-C. (2019).
Trends and development in technology-enhanced
adaptive/personalized learning: A systematic review of
journal publications from 2007 to 2017. Computers &
Education, 140, 103599.
https://doi.org/10.1016/j.compedu.2019.103599
Yarnall, L., Means, B., & Wetzel, T. (2016). Lessons
learned from early implementations of adaptive
courseware. Menlo Park, CA: SRI Education.
Yudelson, M. V., Koedinger, K. R., & Gordon, G. J. (2013,
July). Individualized bayesian knowledge tracing
models. In International conference on artificial
intelligence in education (pp. 171-180). Springer,
Berlin, Heidelberg.
CSEDU 2020 - 12th International Conference on Computer Supported Education
124