Assessment and comparison of meta-features for educational chatbots
data and survey data
DIJANA OREŠKI, DIJANA PLANTAK VUKOVAC, GORAN HAJDIN
Faculty of Organization and Informatics
University of Zagreb
Pavlinska 2, Varazdin
CROATIA
Abstract: - Usage of chatbot platforms is acquiring great attention at all levels of education. Human-chatbot
interactions generate huge amounts of data which are a valuable source of information, when properly analyzed
by means of data and text mining. One of the most challenging tasks in the mining process is the selection of
the appropriate algorithm for a data set at hand. This is a complex task and depends on characteristics of the
dataset used in the analysis. Those characteristics are formalized through meta-features. In this paper, we
identified meta-features of chatbot and survey data. As a case study, we evaluated two data sets and identified
their general meta-features along with discussion. This is, to the best of our knowledge, the first examination of
meta-features for chatbot interactions data and their comparison with survey data.
Key-Words: - meta-features, data characteristics, chatbot interactions, EDUBOTS, educational chatbot, survey
data, meta-learning
Received: June 9, 2021. Revised: January 7, 2022. Accepted: February 17, 2022. Published: March 23, 2022.
1 Introduction
During the past few years, chatbots, which engage
users in conversation to find out their opinions, have
been adopted for a wide variety of applications [1].
Among various chatbot applications, a promising
one is conducting interviews with students. Chatbots
serve as a tool for giving feedback to the students
[2], and have been also used as a new channel of
collecting feedback from students. They are less
time and resource-demanding from traditional
surveys [3]. So far surveys have been widely used,
although their application has several limitations
such as low data quality in the open-ended questions
([4]; [5]). To overcome these limitations individual
interviews are used to gain deeper insights [6].
Recently, chatbot is receiving attention among
practitioners and researchers as a potentially
valuable tool comprising advantages from
qualitative and quantitative evaluations. Their
capability of communication with users through
natural language interaction interfaces serve as
excellent basis to overcome challenges [7]. There is
a small number of papers focusing on the education
[8], especially research papers researching chatbot
usage on students' evaluation of the course ([9],
[10], [3]).
Inspired by these efforts, we are taking a step
forward, and providing description and
characterization of chatbot data. First step in the
characterization process is to identify meta-features
for meta-learning. So far, meta-learning has been
used in general, on publicly available datasets. Most
of the research papers are focused on the analytical
system design, experimental methods or survey
methods [11]. There are only a small number of
papers (e.g. [12]) tackling the educational domain.
Among that, we have not found any papers that
tackle meta-learning and meta-features of chatbot
data within the educational domain. Given the
challenges mentioned above, we focus on the
chatbot data meta-features and their comparison
with traditional survey data. Our investigation has a
goal to discover if there are differences in general
meta-features of chatbot interactions data and
traditional survey data. To achieve that goal, we
evaluated both: chatbots` data with 82 participants
and survey data with 50 participants, both from
University of Zagreb, Faculty of Organization and
Informatics. Both evaluations were focused on the
students' course evaluation.
The rest of the paper is organized as follows. In
Section 2, data is explained and the process of
collecting both chatbot and survey data. In Section
3, an overview of meta-learning is given. The meta-
features are defined and explained. Section 4
WSEAS TRANSACTIONS on ADVANCES in ENGINEERING EDUCATION
DOI: 10.37394/232010.2022.19.3
Dijana Oreški, Dijana Plantak Vukovac,
Goran Hajdin
E-ISSN: 2224-3410
22
Volume 19, 2022
Section 5 summarizes our work, outlines directions
for future research and ends in discussing how meta-
features
2 Literature review
Our work is related to research in three areas:
chatbots in higher education, survey data in higher
education and meta-learning in data mining. All are
explained in the following three subsections, with
the focus of the paper on the first one.
2.1 About chatbots and chatbot data
Depending on the main use of the chatbot data can
be stored in different ways. Most chatbots have a
database which provides the basis for their output.
Educational chatbots are still in an early phase of
artificial intelligence teaching assistants and thus
provide different ways of providing answers to the
users, mostly students [13]. Some chatbots use
predefined entries, while others employ keyword
recognition, closed type decision trees or simple
database searching and matching algorithms [14].
While in the public domain some rely on an open
data approach, others are focused on privacy and
confidentiality, thus using proprietary code [15].
Chatbots are used in different fields, and education
is just a small fraction of their general use. When
considering all fields, the most common chatbot
type in published research papers is CALM-
Systems, followed by Mobile Chatbot, FIUTEBOT
and NDLtutor [11].
Chatbots rely on their data but can also be trained by
different datasets to improve their effectiveness [8].
Some chatbots employ text-mining techniques to
increase interpretation quality of user inputs.
Additionally, they can use event sequence analysis
to further increase the quality of the interpretation
[16].
Some results indicate that chatbots which apply
humanization techniques provide higher data quality
in the context of self-disclosure and social
desirability bias [17]. In the higher education
context chatbots should have a character, consistent
responses, delve into a topic and have some
understanding of factors related to a culture [18].
Most of the papers do not describe chatbot data in
detail, nor do they provide detailed information
about its structure or storing techniques.
2.2 Chatbot and survey data comparisons
Literature review indicated few approaches to
examination of chatbot and survey data. Celino and
Calegari [19] investigated the effectiveness of a
conversational survey in comparison to a traditional
questionnaire. Their results showed that users prefer
conversational form over a traditional approach and
that, from a data collection point of view, the
conversational method shows higher response
quality with respect to a traditional questionnaire.
Yet another perspective is informativeness and
clarity of the responses. Xiao et. al. [20] performed
research of quality responses measured by Gricean
Maxims in terms of their informativeness,
relevance, specificity, and clarity. Their conclusions
indicated a high level of participant engagement
when applying for a chatbot survey. However, they
indicated several drawbacks and provided
guidelines for creating AI-powered chatbots to
conduct effective surveys. Athreya, Ngonga Ngomo
and Usbeck [14] introduced the DBpedia Chatbot, a
knowledge chatbot developed to optimize
interaction. The bot was designed to facilitate the
answering of recurrent questions.
Recent paper of Rhim et. al [17] introduced
humanization survey chatbot, which is another level
of improvement. Authors compared how applying
humanization techniques to survey chatbots can
affect survey-taking experience in three aspects:
respondents' perceptions of chatbots, interaction
experience, and data quality. Regarding data quality,
authors reported better results in the terms of self-
disclosure. Te Pas et al. [21] also compared the user
experience of a chatbot questionnaire with a
traditional questionnaire.
Literature review showed comparisons of chatbot
and traditional survey from different perspectives:
response rate, informativeness or relevance.
However, we did not find any paper tackling this
issue from the perspective of data quality measured
and characterized by meta-features.
2.3 Meta-learning and meta-features
Meta-learning is the process of learning from
previous experience gained during applying various
learning algorithms on different kinds of data and
hence reducing the needed time to learn new tasks
[22]. Main idea of meta-learning is to exploit the
knowledge gained out of previous data analysis
experience [23] and to use this experience of
previous experiments to learn how to improve
automatic learning and recommendation of
algorithms. Meta-learning consists of three steps: (i)
to establish a meta-learning space using meta-data
comprising of meta-features and a performance
measure (meta-response) for machine learning
mining algorithms particular datasets [23], (ii)
developing meta-model out of the meta-dataset
constructed in the first phase, (iii) predictive meta-
WSEAS TRANSACTIONS on ADVANCES in ENGINEERING EDUCATION
DOI: 10.37394/232010.2022.19.3
Dijana Oreški, Dijana Plantak Vukovac,
Goran Hajdin
E-ISSN: 2224-3410
23
Volume 19, 2022
model from second step is used to predict the
performance of an algorithm.
Meta-learning success depends on the input, meta-
features used to describe the given problem. Finding
appropriate meta-features which explain specific
tasks well is a basic problem of meta-learning [24].
Vanschoren [25] 40 meta-features grouped into six
categories: simple, statistical, information, model-
based, and landmarkers. Simple measures are
commonly known and easily extracted from data
[26]. They are also called general measures [24].
Those measures are [25]: Number of instances,
Number of features, Number of classes, Number of
missing values, Number of outliers. Statistical meta-
features give information about data distribution:
average, standard deviation, correlation, and
kurtosis. Statistical measures are used only for
numerical attributes Those measures are:
Correlation, Covariance, Concentration, Skewness,
Kurtosis, ANOVA p-value, Coefficient of variation,
Sparsity, Gravity, PCA 1, PCA skewness, PCA
95%, Class probability. Information meta-features
are from information theory. Information measures
are based on entropy, and they are used for
categorical attributes. Those measures are Mutual
information, Class entropy, Uncertainty coefficient,
Equivalent number of features, and Noise-signal
ratio. Model based features and landmarkers are
specific groups of meta-features which depend on
the modeling algorithm used in the data analysis.
Since this paper is focused on data characterization,
and modeling is not performed, those two groups
are not subject of the investigation in this paper.
3 This research
3.1 Research design: research questions and
methods
Aim of this paper is to research meta-features of
chatbot and survey data. Research design is
constructed around the first three steps of the CRISP
DM process for data mining. CRISP DM consists of
six steps: (i) domain understanding, (ii) data
understanding, (iii) data preparation, (iv) modeling,
(v) evaluation, (vi) deployment. Domain
understanding is described through literature review
in section 2. Data understanding and data
preparation are presented through research results in
section 3. Modeling, evaluation and deployment,
refers to model development and usage of the results
in real-world scenarios, and those steps are not part
of this research.
By using above described methodology and data,
following research questions are addressed:
RQ1: What are general and information-based meta-
features of chatbot interactions data?
RQ2: What are the differences in general and
information-based meta-features between chatbot
interactions data and traditional survey data?
3.2 Research data: chatbot data from the
EDUBOTS project and the faculty's survey
In order to answer to the aforementioned research
questions, we compared data from two different
sources: i) a chatbot Hubert which was investigated
within the project EDUBOTS [27], and ii) a
faculty's survey which is conducted each semester at
the Faculty of Organization and Informatics from
the University of Zagreb [28].
One of the goals of the project EDUBOTS (Best
practices of pedagogical chatbots in higher
education) was to document best practices for the
use of chatbots in higher education, by introducing
two chatbots into university courses: a chatbot BO
within the chat application Differ [29] and a chatbot
Hubert [30], a web application with integrated AI
algorithms developed to automate recruiting process
but also for gathering opinions in the form of the
survey with open-ended questions.
In our research Hubert was used to provide students
with more responsive and entertaining means for
collecting their opinions about the university course.
Educators at three courses created an evaluation for
their course by using a customized educational
template in Hubert. The template consisted of
several main questions that were asking the
following: I) What is working well on
(course_name) and should it continue in the same
way? II) What could teachers start doing, that would
improve it? III) What could teachers stop doing, that
would improve (course_name)? IV) What is your
overall experience of (course_name)? Please write a
sentence or two. V) Do you want to add something
more [3].
Students were provided the link to the course
evaluation in Hubert and were asked to respond in
the form of the sentence in English language.
Subquestions were asked from the Hubert side if it
did not understand the student's answer (Figure 1).
WSEAS TRANSACTIONS on ADVANCES in ENGINEERING EDUCATION
DOI: 10.37394/232010.2022.19.3
Dijana Oreški, Dijana Plantak Vukovac,
Goran Hajdin
E-ISSN: 2224-3410
24
Volume 19, 2022
Figure 1. Conversation in Hubert for course
evaluation
All conversations related to one course could be
downloaded in the form of a transcript in a .txt file.
In addition, collected answers are visualized in a
Hubert evaluation dashboard, in the form of
numerical data and graphs. Data are also classified
according to the positive or negative tone of the
answer (see Figure 2).
Figure 2. Data visualisation in Hubert
The second set of data is from the faculty's survey
about perceptions of the students about the course
quality and its delivery in the online environment.
This survey was introduced in summer semester
2019/2020 after all teaching was shifted to the full
online in both asynchronous mode (Moodle, video
lectures etc.) and synchronous mode (e.g.
videoconferences and chats with the students) due to
COVID-19 pandemic. The survey is now used
regularly in each semester to evaluate every faculty
course and its teaching quality, to identify trends
and room for improvements.
The survey consists of 27 questions grouped into the
following categories: I) course organization and
communication, II) teaching materials on LMS, III)
knowledge and skills assessment, and IV) delivery
of the course. Students rated their opinions about the
course on the Likert scale from 1 (totally disagree)
to 5 (totally agree) or selected a single or multiple
response from a predefined list, but also had the
opportunity to provide the answers to four open-
ended questions. Figure 3 shows the example of the
data visualization of the answers to two questions
from the category Course organization and
communication.
Figure 3. An example of data visualisation in FOI
Students surveys
In the section 4, research results are presented with
an aim to answer research questions.
WSEAS TRANSACTIONS on ADVANCES in ENGINEERING EDUCATION
DOI: 10.37394/232010.2022.19.3
Dijana Oreški, Dijana Plantak Vukovac,
Goran Hajdin
E-ISSN: 2224-3410
25
Volume 19, 2022
4 Results
Meta-learning process consists of the: (i) data
collection, (ii) data understanding activities which
are focused at detecting data quality. Two datasets
are presented in this paper: one collected with the
chatbot Hubert within the EDUBOTS project and
the second one collected with the classical survey.
First dataset consists of chatbots` data with 82
participants, students which were asked to evaluate
the course. Second dataset consists of survey data
with 50 participants, students which were asked to
evaluate the course through a survey questionnaire.
Research was conducted at the winter semester of
2020/2021 academic year (survey data) and summer
semester (chatbot data) of 2019/2020 academic
year. Chatbot data was collected within the
following three undergraduate courses at University
of Zagreb, Faculty of Organization and Informatics:
Software Engineering, Text and Image Editing and
Business Informatics. Via forum messages on
Moodle, students were asked to give feedback and
evaluate the course using chatbot application
Hubert.
Survey data was collected within the following two
courses at the undergraduate level: Knowledge
based systems and Knowledge discovery in data,
and one course at the graduate level: Intelligent
systems.
On the previously explained data sets, meta-features
are extracted. General meta-features include general
information related to the dataset at hand. To a
certain extent they are conceived to measure the
complexity of the underlying problem. Some of
them are the number of instances, the number of
attributes, dataset dimensionality, the ratio of
missing values. Table 1. depicts main meta-features
for both datasets.
Table 1. General meta-features for chatbot and
survey data
Meta-feature
Chatbot data
Survey data
Number of
instances
82
50
Number of
attributes
7
17
Number of
categorical
attributes
7
17
Number of
numerical
attributes
0
0
Ratio
categorical to
numerical
0/7
0/17
15 %
22 %
Statistical meta-features are defined for numerical
attributes since those meta-features describe
attribute statistics and class distributions of a dataset
sample. They include various summary statistics per
attribute like mean, standard deviation, class
entropy, etc. Hereinafter, information based meta-
features are calculated since those features are
intended for categorical attributes, and two datasets
involved in the research consists of categorical
attributes
Table 2. Information based meta-features for
chatbot and survey data
Chatbot data
Survey data
0.43
0.66
0.57
0.34
0.26
0.33
1
1
0.17
0.33
According to information based meta-features,
chatbot data gives more relevant information
(measured by mutual information) and gives less
noise (measured by noise-signal ratio).
5 Discussion
In this section, we provide answers on research
questions and discuss our results. General meta-
features of chatbot data are: low number of
instances and low number of attributes, higher
number of categorical attributes than numerical, and
low number of missing values (see Table 1, RQ1).
In our sample data, both sets had low number of
instances and low number of attributes. Information-
based meta-features of chatbot data are low level of
mutual information, low level of uncertainty
coefficient and noise-signal ratio (see Table 1,
RQ1). Comparison of chatbot interactions data and
traditional survey data yielded differences in
number of attributes, number of instances, number
of missing values, class entropy, mutual information
and noise-signal ratio (see Table 2, RQ2)
WSEAS TRANSACTIONS on ADVANCES in ENGINEERING EDUCATION
DOI: 10.37394/232010.2022.19.3
Dijana Oreški, Dijana Plantak Vukovac,
Goran Hajdin
E-ISSN: 2224-3410
26
Volume 19, 2022
In this research, we performed data characterization
through meta-features for both human-chatbot
interaction and survey data and provided their
comparison. Our approach focuses on general and
information meta-features with an aim to detect data
quality. General and information meta-features
show higher data quality for chatbot data. In order to
give a broader conclusion, this paper investigated
previous comparisons of chatbot and survey data,
from various perspectives. Some of the previous
research papers on the topic are presented in the
second section of the paper. Our results are in line
with previous research papers investigating a given
topic, but from different perspectives than ours.
Celino and Calegari [19] reported higher response
quality of chatbot evaluations with respect to a
traditional questionnaire. Furthermore, Rhim et. al
[17] showed better data quality of chatbot data
measured through self-disclosure. Our investigation
showed that chatbot data gives more relevant
information (measured by mutual information) and
gives less noise (measured by noise-signal ratio).
Implications of such results provide valuable
contributions. Results indicate that chatbots can
serve as valuable and quality tool for data
collection.
6 Conclusion
Despite recent advances in machine learning, it is
still challenging to find appropriate algorithms for
data analysis. Especially, with a growing body of
data sources emerging every day. Selection of
appropriate algorithms is dependent on the meta-
features of employed data. Thus, meta-features
should be explored and elaborated to characterize a
specific domain. In this research, we focused on
chatbot data and compared it with traditional
sources in the educational domain in order to
understand characteristics of datasets. One group of
meta-features, general meta-features, were
investigated to understand data properties. Our
research resulted with the following scientific
contributions: (i) identification of meta-features in
chatbot data, (ii) comparison of chatbot data meta-
features with survey data meta-features, (iii) Since
this is only the first part of the research, it has
several limitations. We have employed only general
and information meta-features into account.
Secondly, only two datasets were included in the
research. In the future research, there will be a
broader number of chatbot datasets to increase the
possibility of results generalization. Furthermore,
other groups of meta-features will be employed in
order to make a reliable base for meta-model
development.
References:
[1] Grech, M. 2017. The Current State of
Chatbots in 2017. (Apr 2017).
https://getvoip.com/blog/2017/04/21/the-
current- state- of- chatbots- in- 2017/
[2] Lundqvist, K. O., Pursey, G., & Williams,
S. (2013). Design and implementation of
conversational agents for harvesting
feedback in eLearning systems. In
European Conference on Technology
Enhanced Learning (pp. 617-618).
Springer, Berlin, Heidelberg.
[3] Čižmešija, A., Hajdin, G., Oreški, D., Using
chatbot for course evaluation in higher
education, INTED2021 Proceedings, 1494-
1501.
[4] Erikson, M., Erikson, M. G., and Punzi, E.
(2016). Student Responses to a Reflexive
Course Evaluation. Reflective Practice
(17:6), Routledge, pp. 663–675.
(https://doi.org/10.1080/14623943.2016.120
6877).
[5] Tucker, B., Jones, S., and Straker, L.
(2008). Online Student Evaluation Improves
Course Experience Questionnaire Results in
a Physiotherapy Program. Higher Education
Research and Development (27:3), pp. 281–
296.
(https://doi.org/10.1080/0729436080225906
7).
[6] Steyn, C., Davies, C., and Sambo, A.
(2019). Eliciting Student Feedback for
Course Development: The Application of a
Qualitative Course Evaluation Tool among
Business Research Students. Assessment
and Evaluation in Higher Education (44:1),
Routledge, pp. 11–24.
(https://doi.org/10.1080/02602938.2018.146
6266).
[7] Rubin, V. L., Chen, Y., and Thorimbert, L.
M. (2010). Artificially Intelligent
Conversational Agents in Libraries. Library
Hi Tech (28:4), pp. 496–522.
(https://doi.org/10.1108/0737883101109619
6
[8] Baha, T. A. I. T., Hajji, M. E. L., Es-Saady,
Y., & Fadili, H. (2022). Towards highly
adaptive Edu-Chatbot. Procedia Computer
Science, 198, 397–403.
[9] Wambsganss, T., Winkler, R., Schmid, P.,
& Söllner, M. (2020a). Unleashing the
WSEAS TRANSACTIONS on ADVANCES in ENGINEERING EDUCATION
DOI: 10.37394/232010.2022.19.3
Dijana Oreški, Dijana Plantak Vukovac,
Goran Hajdin
E-ISSN: 2224-3410
27
Volume 19, 2022
Potential of Conversational Agents for
Course Evaluations: Empirical Insights
from a Comparison with Web Surveys. In
ECIS.
[10] Wambsganss, T., Winkler, R.,
Söllner, M., & Leimeister, J. M. (2020b). A
Conversational Agent to Improve Response
Quality in Course Evaluations. In Extended
Abstracts of the 2020 CHI Conference on
Human Factors in Computing Systems (pp.
1-9).
[11] Wang, J., Hwang, G. H., & Chang,
C. Y. (2021). Directions of the 100 most
cited chatbot-related human behavior
research: A review of academic
publications. Computers and Education:
Artificial Intelligence, 100023.
[12] Molina, M.M., Romero, C. Luna,
J.M. Ventura. S. Metalearning approach for
automatic parameter tuning: A case study
with educational datasets. 5th International
Conference on Educational Data Mining,
Chania, Greece,180-183, 2012.
[13] Smutny, P., & Schreiberova, P.
(2020). Chatbots for learning: A review of
educational chatbots for the Facebook
Messenger. Computers & Education, 151,
103862.
[14] Athreya, R. G., Ngonga Ngomo, A.
C., & Usbeck, R. (2018, April). Enhancing
Community Interactions with Data-Driven
Chatbots--The DBpedia Chatbot. In
Companion Proceedings of the The Web
Conference 2018 (pp. 143-146).
[15] Keyner, S., Savenkov, V., &
Vakulenko, S. (2019, June). Open data
chatbot. In European Semantic Web
Conference (pp. 111-115). Springer, Cham.
[16] Akhtar, M., Neidhardt, J., &
Werthner, H. (2019, July). The potential of
chatbots: analysis of chatbot conversations.
In 2019 IEEE 21st Conference on Business
Informatics (CBI) (Vol. 1, pp. 397-404).
IEEE.
[17] Rhim, J., Kwak, M., Gong, Y., &
Gweon, G. (2022). Application of
humanization to survey chatbots: Change in
chatbot perception, interaction experience,
and survey data quality. Computers in
Human Behavior, 126, 107034.
[18] Tsivitanidou, O., & Ioannou, A.
(2020). Users' Needs Assessment for
Chatbots' Use in Higher Education. In
Central European Conference on
Information and Intelligent Systems (pp. 55-
62). Faculty of Organization and
Informatics Varazdin.
[19] Celino, I., & Calegari, G. R. (2020).
Submitting surveys via a conversational
interface: an evaluation of user acceptance
and approach effectiveness. International
Journal of Human-Computer Studies, 139,
102410.
[20] Xiao, Z., Zhou, M. X., Liao, Q. V.,
Mark, G., Chi, C., Chen, W., & Yang, H.
(2020). Tell me about yourself: Using an
AI-powered chatbot to conduct
conversational surveys with open-ended
questions. ACM Transactions on Computer-
Human Interaction (TOCHI), 27(3), 1-37.
[21] Te Pas, M. E., Rutten, W. G.,
Bouwman, R. A., & Buise, M. P. (2020).
User experience of a chatbot questionnaire
versus a regular computer questionnaire:
prospective comparative study. JMIR
Medical Informatics, 8(12), e21982.
[22] Brazdil, P., Giraud Carrier, C.,
Soares, C., Vilalta, R. (2009). Metalearning:
Concepts and Systems (pp. 1–10).
https://doi.org/10.1007/978-3-540-73263-
1_1
[23] Bilalli, B., Abelló, A., & Aluja-
Banet, T. (2017). On the predictive power
of meta-features in OpenML. International
Journal of Applied Mathematics and
Computer Science, 27(4), 697–712.
https://doi.org/10.1515/amcs-2017-0048
[24] Castiello, C., Castellano, G., &
Fanelli, A. M. (2005). Meta-data:
Characterization of input features for meta-
learning. In Lecture Notes in Computer
Science (including subseries Lecture Notes
in Artificial Intelligence and Lecture Notes
in Bioinformatics): Vol. 3558 LNAI (pp.
457–468).
https://doi.org/10.1007/11526018_45
[25] Vanschoren, J. (2018). Meta-
Learning: A Survey.
http://arxiv.org/abs/1810.03548
[26] Rivolli, A., Garcia, L. P. F., Soares,
C., Vanschoren, J., & de Carvalho, A. C. P.
L. F. (2018). Characterizing classification
datasets: a study of meta-features for meta-
learning. http://arxiv.org/abs/1808.10406
[27] EDUBOTS. (2021, January 8)
Realizing Chatbots in Higher Education.
Retrieved from: www.edubots.eu
[28] FOI Students surveys. (2021) FOI
studentske ankete. Retrieved from:
https://ankete.foi.hr/
WSEAS TRANSACTIONS on ADVANCES in ENGINEERING EDUCATION
DOI: 10.37394/232010.2022.19.3
Dijana Oreški, Dijana Plantak Vukovac,
Goran Hajdin
E-ISSN: 2224-3410
28
Volume 19, 2022
[29] Differ (2021, January 8) For
Educators. Retrieved from:
www.differ.chat/for-educators
[30] Hubert (2021, October) Hubert.ai,
Retrieved from: https://www.hubert.ai/
Contribution of individual authors to
the creation of a scientific article
(ghostwriting policy)
Dijana Oreški, Dijana Plantak Vukovac and Goran
Hajdin collected data, prepared data and analysed
data.
Dijana Oreški carried out interpretation of results.
Dijana Plantak Vukovac and Goran Hajdin
organized literature review.
Sources of funding for research
presented in a scientific article or
scientific article itself
This work has been supported by the project “Best
practices of pedagogical chatbots in higher
education / EDUBOTS” which is funded under the
Erasmus+ KA2: Cooperation for innovation and the
exchange of good practices – Knowledge Alliances
(grant agreement no: 612446, project ref.: 612446-
EPP-1–2019-1-NO-EPPKA2-KA).
Creative Commons Attribution
License 4.0 (Attribution 4.0
International , CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en
_US
WSEAS TRANSACTIONS on ADVANCES in ENGINEERING EDUCATION
DOI: 10.37394/232010.2022.19.3
Dijana Oreški, Dijana Plantak Vukovac,
Goran Hajdin
E-ISSN: 2224-3410
29
Volume 19, 2022