Conversational Analysis to Recommend Collaborative Learning in

Distance Education

Antônio J. Moraes Neto

, Márcia A. Fernandes

and Tel Amiel

Federal Institute of Brasilia (IFB), Brasília, DF, Brazil

Computer Science Graduate Program, Federal University of Uberlândia (UFU), Uberlândia, MG, Brazil

Education Graduate Program, University of Brasilia (UnB), Brasília, DF, Brazil

Keywords: Conversational Agent, Computer-supported Collaborative Learning, Collaborative Learning Assessment,

Artificial Intelligence in Education, Educational Technology.

Abstract: Conversational agents can recommend interactions among students in a Virtual Learning Environment (VLE)

for the purpose of supporting collaborative learning, an important approach to improve online education. This

paper describes the current position of a research that addresses the implementation of Conversational

Analysis (CA) in order to make recommendations through chatbots for promoting collaborative learning

among students in a VLE. Based on an experiment, the authors propose a CA strategy to determine the level

of collaboration among students, point out possibilities for chatbot’s intervention in favor of collaborative

learning, and present the results obtained in the current stage of the research.

1 INTRODUCTION

Conversational Analysis (CA) offers a way to analyze

the understanding produced through interaction,

focusing on the methods by which interactants build

sense collaboratively, with the aim of producing a

report on how understanding was achieved in the

conversation (Koschmann, 2013). A CA

methodological approach can assess not only the

content, but also the structure, nature of roles, and

relationships within students’ conversations

(Abraham et al., 2016).

The characterization of the online conversation

provided by CA can be used in Virtual Learning

Environments (VLEs) in order to better identify

where social interaction occurs and how it takes

place, indicating possibilities for collaboration. It can

also help indicate where no collaboration has taken

place, but possibilities exist for interactions and to

promote collaborative learning, in which students do

not depend only on direct interaction with the content

and teachers since the possibilities are expanded

through the student-student connection. Thus, they

learn through their doubts and interests, teaching each

https://orcid.org/0000-0002-6139-8919

https://orcid.org/0000-0003-3572-612X

https://orcid.org/0000-0002-1775-1148

other. At the same time, they can visualize how others

are learning as well as their difficulties, which

demands a computational support oriented to

productive interaction in a motivating way (Stahl et

al., 2005).

Chatbot as a pedagogical tool offers opportunities

to support learning in adaptive and personalized

environments (Zawacki-Richter et al., 2019). For

example, a chatbot integrated into a VLE can provide

predefined feedback during the chat in order to

intervene and encourage students’ engagement in the

conversation, and keep focus on one aspect of the task

at hand (Tawfik et al., 2020). Therefore, chatbots can

be used to instigate debate among students in a VLE

and, in case there is interaction or the absence of it,

indicate to the teacher where the collaboration is

occurring or could occur. In addition, chatbots can

suggest actions to stimulate collaborative learning.

Considering this potential for the application of

Artificial Intelligence in Education (AIEd),

establishing a method to measure collaboration

among students is relevant. In this paper this authors

discuss a methodology to measure collaboration

levels based on what students write in discussion

196

Moraes Neto, A., Fernandes, M. and Amiel, T.

Conversational Analysis to Recommend Collaborative Learning in Distance Education.

DOI: 10.5220/0011092600003182

In Proceedings of the 14th International Conference on Computer Supported Education (CSEDU 2022) - Volume 2, pages 196-203

ISBN: 978-989-758-562-3; ISSN: 2184-5026

forums, so as to make recommendations to students,

teachers, and tutors in order to promote collaborative

learning in VLEs.

2 RESEARCH AIMS

The main objective of the research is the

implementation of a chatbot adopting CA to make

recommendations in order to promote collaborative

learning in a VLE. The specific objectives are: (1) to

analyze aspects of students’ interactions that may

indicate the collaboration among them in the

discussion forums appointed by the teacher as those

which are to be monitored; (2) to model knowledge

on the level of collaboration among students in the

monitored forums; (3) to model the dialogue base in

an available chatbot architecture, that can be

integrated into a VLE, to structure the conversations

with students, teachers, and tutors; and (4) to make

recommendations to the participants, focusing on

questions that encourage feedback and on topics

under discussion, in order to promote the

collaboration among students in the monitored

discussion forums.

The research hypothesis is that adopting CA with

a chatbot makes it possible to make recommendations

to students, teachers, and tutors, in order to promote

collaborative learning in distance education. The

following research questions are associated with this

hypothesis: (1) how can one characterize students’

interactions through a CA in monitored discussion

forums?; (2) what knowledge can be modeled in

regards to the level of collaboration among students

in monitored discussion forums?; (3) in which

architecture that can be integrated into a VLE is it

possible to model the chatbot dialogue with students,

teachers, and tutors?; and (4) what recommendations

should the chatbot make to students, teachers, and

tutors, focusing on questions that encourage feedback

and on topics under discussion, to promote the

collaboration among students in monitored

discussion forums?

Moodle

, an acronym for Modular Object-

Oriented Dynamic Learning Environment, is a free

software VLE and was chosen as the ET development

platform, as it is the most prevalent VLE in higher-

education institutions in the Portuguese language

context. In this environment, a teacher will define

which forums will be monitored. Moreover, the ET

processing will be done only once for each post made

in a monitored discussion forum.

Moodle website: https://moodle.org/?lang=en.

3 LITERATURE REVIEW

The literature review were carried out both by

focusing on conceptual aspects and methodological

emphasis. The first part relates to chatbots,

collaborative learning, and CA. The results indicated

that CA to identify collaboration in VLE and

possibilities of chatbot’s intervention in favor of

collaborative learning is a promising research area for

AIEd.

CA offers a way to analyze the understanding

produced through interaction, focusing on the

methods by which interactants build sense

collaboratively. It can also provide insight on how

understanding was achieved in the conversation

(Koschmann, 2013). CA focuses on the sequential

nature of thinking, which is lost in most statistical

coding analyses, where individual statements are

encoded and then accounted for, without regard to

their sequential response order (Stahl, 2012). The

adoption of CA is relevant when considering the

characteristics of online conversations, whose

grammar is comparatively informal and unstructured,

with users involved in a tone of conversation,

compared to other texts (Uthus & Aha, 2013).

From the analysis of conversation logs, it is

possible to adopt a methodology to detect and classify

student interaction behavior (Procter et al., 2018). In

order to assess collaboration among students, it is

necessary to use interaction analysis techniques that

identify some of the processes used by groups to

create meaning and build knowledge, providing an

insight into collaboration according to the sequential

flow of students’ statements. As students are solving

problems together, they necessarily express their

thoughts to each other and this data is available for

analysis in VLE logs. Moreover, the flow of

proposals, responses, questions, agreements, etc. is

available for analysis as an extended cognitive

process (Stahl, 2012). In order to analyze this data it

is necessary to adopt preprocessing practices that

avoid overly optimistic results in the analysis of

discussion forums (Farrow et al., 2019).

Some aspects related to the assessment of

collaboration among students deserve to be

highlighted in the context of this research. In the

literature on classroom discourse, the adjacent pair

becomes a ’utterance-triad’, question-answer-

comment, which is commonly described as the

sequence IRE (Inquiry, Response, Evaluation), the

latter referring to the sufficiency of that answer. This

indicates that the basic and minimum form of a

Conversational Analysis to Recommend Collaborative Learning in Distance Education

197

sequence is two turns of conversation and that the

sequences composed of more turns are expansions,

which can produce an assessment of conversation,

positive or not, in the third round (Koschmann, 2013).

To identify questioning, Lu et al. (2011) propose that

this is a type of statement that seeks factual

information, including words such as “what”,

“which”, “where” and “when”, or one that seeks

explanation, including words such as “why” and

“how”. To identify questions, the Linguistic Inquiry

and Word Count (LIWC) software package, which is

based on empirical research, can be used to extract

word counts indicative of different psychological

processes, such as affective, cognitive, social and

perceptual (Farrow et al., 2019). Its core is based on

a lexical resource, called the LIWC dictionary, which

is also available in Portuguese (Cavalcanti et al.,

2020).

The quality of engagement in educational tasks is

measured by the number of responses to posts, and

not by the number of posts initiated by an individual

student, that is, responses demonstrate engagement

(Lyndall & Elspeth, 2015). The number of debating

students also influences the quality of their

interactions, ideally being organized in small groups,

ranging from 3 to 6 participants, which positively

impacts the value of the discussions (Saqr et al.,

2019). Social Network Analysis (SNA) makes it

possible to record the number of interactions among

students as an indicator of quality in collaboration.

The use of SNA has played a prominent role in the

analysis of learning in order to indicate collaborative

learning (Dascalu et al., 2018). It is also important to

note that the benefit of measuring the quality of

collaboration for individual students is the

recognition of their proactive and effective

collaboration (Lyndall & Elspeth, 2015).

Regarding topic detection, the repetition of

keywords in statements by different students is an

indicator of which topics are under discussion

(Allaymoun & Trausan-Matu, 2015). To this end,

topic modeling, a text mining tool frequently used to

discovery hidden semantic structures in a corpus, can

be adopted to identify keywords in student

statements. Based on this identification, Epistemic

Network Analysis (ENA) combined with SNA can

detect information about the student performance in

the perspective of identifying a set of cognitive and

social dimensions, which is marked by interaction

with the appropriate people on the appropriate content

(Farrow et al., 2019).

Some collaborative learning factors relevant to

chatbot performance are characterized regarding the

effectiveness of immediate feedback, more

appropriate in verbal learning tasks, and delayed

feedback, advantageous in learning concepts because

it allows more time for students’ metacognition;

being careful not to interrupt or disturb when there are

interactions among students during their learning

activities; and the benefit more focused on

interactions among students than on their learning

performance (Hayashi, 2019).

Hayashi (2019) implemented the following three-

steps chatbot structure: (1) two chatbots were

designed to facilitate requests based on types of

functions: the communication consultant to answer

about the efficiency of communication and the tutor

of explanations to generate answers on how to think

about a topic that triggers metacognition; (2) the

system detected keywords in an inserted sentence and

classifies them by type; (3) the system generates

responses based on detected keywords and number of

turns taken in conversation. Each chatbot, therefore,

responded to students when it detected any of the

keywords, whether they are related to important

phrases or communication problems (Hayashi, 2019).

Classification processes have been implemented

through machine learning algorithms, which is a sub-

field of AI capable of recognizing patterns, making

predictions and applying newly discovered patterns in

situations that were not initially included or covered.

Zawacki-Richter et al. (2019) identified, in a review

of 58 studies in this area, that all of them applied

machine learning methods to recognize and classify

patterns and model student profiles. To evaluate the

accuracy of classifiers, the authors used statistical

measures that demonstrated their high ability to

predict the performance in a student group from

participating in online discussion.

With regard to recommendation systems,

Chatbots can play an effective role in distance

education, having been identified as an ET that may

contribute to the acceleration of the learning process,

facilitate access to educational contents and enrich the

learning environment by supporting students and

teachers (Liu et al., 2019). It is also relevant to

highlight that in knowledge-based recommendation

systems, recommendations are suggested based on

the specified requirements, and not on the learner’s

interaction history (Aggarwal, 2016).

Chatbot intervention strategies can be defined

based on the Academically Productive Talk (APT)

structure, designed to encourage discussion in an

educational context from social interaction to the

construction of mental processes, with an emphasis

on valuable interaction (Tegos et al., 2020). APT

proposes tools to be adopted by the teacher in order

to encourage discussion in the classroom in which

CSEDU 2022 - 14th International Conference on Computer Supported Education

198

students expose their reasoning, listen deeply and

critically to the contributions of others, and thus

interact collaboratively (Michaels & O’Connor,

2015). It is also important to note that when students

post their participation in forums and their partners

receive invitations to comment on them, this results

in a smaller number of fragmented topics, but with a

greater number of participations per topic (Oliveira et

al., 2011).

In conclusion, the adoption of a ET in order to

carry out the recommendation of collaborative

learning in distance education starts with the CA to

identify the possibility of collaboration. A chatbot,

whose architecture includes CA and machine

learning, in addition to providing technical

information and educational content, can promote

collaborative learning in VLE through interventions

that contribute to the construction of students’

knowledge. Thus, this application of AIEd can act as

a recommendation system when it is implemented as

a resource that makes feasible the debate among

students, providing knowledge about the domain,

supporting the affective and social experience, and

contributing to the proper usability of VLE, which

can occur even through mobile devices.

4 RESEARCH METHOD

The methodological approach takes place in three

stages: conversational analysis; assessment to

determine collaboration level; and implementation of

the chatbot to make recommendations to students,

teachers, and tutors in the monitored discussion

forums. Accordingly, the development of the ET has

been taking place in the stages shown in Figure 1.

Figure 1: Three stages of the methodological approach.

In the next subsections, the authors present the

results obtained in these stages, not only describing

the conceptual model, but also showing some relevant

aspects for its implementation.

4.1 Conversational Analysis

The adopted CA seeks to identify interactions among

students from the text and metadata of each post

obtained in the monitored discussion forums, whose

implementation is described in the steps below. The

CA layer applied in the context of research’s

architectural flow is shown in Figure 2.

Figure 2: Architectural flow for the research.

For each message posted by a participant, the CA

steps must be performed, whose resulting information

must persist in a relational Database Management

System (DBMS). The set of data was obtained via

SQL, from read-only access to the VLE database of a

vocational education school, from which two online

courses were offered. The posts obtained are

exclusively from the discussion forums, without

participant identification, as only messages and

forums are assessed in order to generate

recommendations. From the available 20,976

messages, 15,703 were posted by students.

Preprocessing is the CA step in which specific

techniques of Natural Language Processing (NLP)

are applied, without which the quality of the results

will be compromised. First, it is important to clean up

the obtained data, such as deleting HTML tags and

punctuation marks used on web addresses, and

formatting numeric fields. Subsequently, NLP

techniques take place, as lemmatization, mapping

inflected forms of word to a common root; stemming,

removing the ending of words to find their base form;

and phonetic mapping that addresses features rarely

seen in formal texts, which can be applied to words

and numbers to define the meaning of words with

unusual spellings. In this research some NLP

Conversational Analysis to Recommend Collaborative Learning in Distance Education

199

techniques are adopted in step Detection and

Tracking of Topics, described below.

Resource processing is the CA step in which the

characterization of social dynamics occurs through

SNA, carried out by the Cytoscape

open source

software platform, in order to identify interactions

among students in each forum. To this end, SNA

provides insights into dimensions such as cohesion,

centralization, and prominence. Centrality measures

seek to identify the extent to which the network

depends on a certain number of interactants. Thus, the

Weighted Degree Centrality (WDC) is responsible

for the weight of the edges that a node has in the

network, being the sum of the edges weights

connected to the node. SNA also provides in- and out-

degree (OD) metrics, which are scores that

correspond to the in-and-out edges of a given node

calculated from the sociogram (Pereira, 2018).

In this research, each node corresponds to a

message in the forum. Therefore, WDC characterizes

the number of student responses, as the weight of

their messages is one and that of the other interactors

is zero. When OD is zero, then there was no response

for the given message, but if it is greater than zero,

then it is because there were that specific number of

responses for the message.

Identification of the message attribute is the CA

step that allows identifying characteristics of the

statements, specifically the questions, through NLP,

using LIWC

. Based on the total words count (WC),

LIWC informs percentages such as Interrog and

QMark, related to question words and question mark

respectively.

This research is developed in a Portuguese

language context, and, considering the results

obtained from the LIWC Portuguese dictionary, only

the presence of the question mark was effective to

identify the questions, that is, it was not possible to

identify any question message without the QMark

percentage was grater than zero.

Topic detection and tracking is the CA step in

which key terms discussed in each forum are

identified through topic modeling, made with the

open source software Tomotopy

, which is a topic

modeling toolkit used as a Python module. Tomotopy

implements one of the earliest and most widely

utilized topic modeling methods called Latent

Dirichlet Allocation (LDA), which defines hidden

topics to capture latent semantics in text documents.

With LDA, each document is represented by a

probability distribution (dirichlet) over topics, which

Cytoscape website: https://cytoscape.org/.

LIWC website: http://liwc.wpengine.com/.

are hidden (latent), with each topic being described

by a distribution over self-explanatory words

(allocation). Thus, the LDA algorithm infers

unobserved topics, which do not contain labels that

would describe them, by assigning words to topics

placing terms that often appear together in a

document, it means, topics are a collection of the

proportions of their contents, where word order is

irrelevant (Schulte, 2021). This machine learning tool

is commonly used in a few areas of focus, including

document classification and recommendation of new

articles that are likely to be of interest to a specific

reader.

Another method implemented by Tomotopy is

called the Correlated Topic Model (CTM), which is

similar to LDA, but it can be used to describe the

latent composition of associated topics in pairs within

each document in a corpus. For LDA and CTM, the

variable K defines the number of topics to be

generated. The parameters φ and θ are seen as mixture

weights and characterize the probability of

importance of words for a given topic and the

proportion of topics within a specific document,

respectively. Thus, the topic modeling algorithm

calculates φ

(z)

to represent the multinomial

distribution of terms over a given topic z, and works

out θ

(d)

to represent the multinomial distribution of

topics about a given document d (Vayansky &

Kumar, 2020).

Within this paper, a “word” or “term” represents

the fundamental unit of data, a “document” represents

a message posted by one participant, and a “corpus”

represents a group of documents encompassing the

entire discussion forum. A “vocabulary” is the

collection of all distinct words within a corpus, and a

“topic” is a probability distribution spanning a given

vocabulary. In this context, the LDA and CTM are

being applied in order to: (1) identify the topic that

has the largest number of words in the corpus

associated with it; (2) classify each message

according to the percentage of words associated with

the identified topic; and (3) point out the most

relevant terms within the corpus aiming to show some

messages to which a student can post their

contribution.

4.2 Determining Collaboration Level

In the present research, collaborative learning must

occur from the interaction among students in a

discussion forum, in which they jointly address one

Tomotopy: https://bab2min.github.io/tomotopy.

CSEDU 2022 - 14th International Conference on Computer Supported Education

200

or more topics, through replies to previous messages,

characterizing a conversation. If there is a question in

any topic under discussion, it is desirable to have a

colleague’s response to this question, which

characterizes an answer. In order to infer the level of

collaboration among students in monitored

discussion forums, it is necessary to:

 Create the initial database from the CA with

real data, collecting information from Moodle

forums for the assessment of collaboration

among students, including indicators to be

evaluated by teachers;

 Assessment by teachers as to which

combinations of the mentioned indicators are

better for classifying collaboration among

students, generating a new database that will

allow to learn, in an automated and intelligent

way, how to classify this type of collaboration;

 Apply machine learning and other techniques

to the database resulting from the evaluation by

teachers in order to carry out the evaluation of

collaboration among students.

The assessment of collaboration based on

interaction is made through CA by the combination

of variables that indicate where it occurred, including

insights from the aforementioned Literature Review,

to compose the following indicators:

 Identification of Students' Interactions (ISI),

performed by Resource Processing, to

characterize the amount of student responses to

each message, which is calculate by the

formula 1 below;

 Questioning Characterization (QC), carried out

by the Message Attribute Identification, to

point out each student message that contains a

question, which is worked out by the formula

 Main Topic Approached (MTA), which occurs

from the Detection and Tracking of Topics,

which aims to infer the topic with the highest

word distribution in each discussion forum,

where MTA is the value of the proportion of

this topic in each message, as shown in the

formula 3;

 Students Collaboration Level (SCL) of each

message is formed by the average of the

previous indicators, as shown in the formula 4.

ISI = WDC / OD for OD greater than zero (1)

QC = QMark / 100 (2)

MTA = θ

(d)

of the highest φ (3)

SCL = ISI + QC + MTA / 3 (4)

In Table 1 are presented results of the CA layer in

a Portuguese Language forum that took place at the

beginning of the second semester of an online course,

containing 47 messages, 31 of which were posted by

students, 2 by the teacher, and 14 are posted by a

tutor.

Table 1: An example of SCL calculation.

QC OD WDC ISI MTA SCL

0.0101 23 23 1.0 0.08369590 NA

0.0303 2 1 0.5 0.08897769 0.20642590

0.0000 2 1 0.5 0.23834153 0.24611384

0.1250 1 0 0.0 0.05000000 0.05833333

For the first message, in Table 1, SCL is equal to

“NA” because the calculation is not applicable for a

teacher post, which in this case was responded to

directly by 23 messages (OD) all posted by students

(WDC). The message corresponding to the second

line got 2 responses, 1 from a student, and therefore

its ISI is equal to 0.5. Its MTA corresponds to the

proportion of the topic with the highest word

distribution among the 20 topics generated by CTM.

There is no question mark (QC) in the student

message on the third line, but it still got a return from

a colleague, probably because it covered the topic

more than in the others posts, as its MTA indicates.

In the message on the last line there is a higher QC,

which can be considered a more specific question by

the student, who addresses the main topic (MTA) a

little, but has not yet received feedback from a

colleague (WDC), but only from the teacher and so

OD is equal to 1.

Thus, the three indicators of collaboration will be

combined, based on real data, considering assessment

of teachers. The database resulting from their

assessment will be constantly updated in order to

adjust the classification of collaboration among

students. It is important to highlight that the

intelligent processing of the mentioned indicators will

occur in order to classify the conversations among

students regarding their SCL. These inferences will

allow chatbot recommendations to be generated for

students, teachers, and tutors in order to promote

collaboration among students.

4.3 Chatbot Recommendations for

Student Collaboration

Chatbots to be implemented, using an existing tool,

will then be able to make recommendations for each

situation from the context identified in the previous

stages. Based on the APT structure, recommendations

to students aim:

Conversational Analysis to Recommend Collaborative Learning in Distance Education

201

 To suggest options for motivating student

participation by prioritizing their messages

with (1) questions that are still unanswered, (2)

main terms under discussion, and (3) student

responses to a colleague;

 To provide information about each monitored

forum the student has participated in, focusing

on (1) number of student messages with

percentages of questions and returns, (2)

indication of message collaboration level, and

(3) main terms discussed.

Recommendations to teachers and tutors aim:

 To suggest options for motivating student

participation, prioritizing messages (1) that

contain questions which are still unanswered;

(2) those messages that least cover the main

terms under discussion; and (3) those messages

that do not yet have responses from colleagues;

 To provide information about monitored

forums, focusing on (1) number of messages

from students with percentages of questions

and returns, (2) amount of recommendations

made, (3) percentage of recommendations that

generated participation, (4) main terms

discussed, and (5) classification of messages

regarding the level of collaboration.

The chatbots will start the dialogue with students

or teachers when they access a discussion forum that

is being monitored. This agents will also send

messages to the participants to specifically inform

them about new conversations to participate in. A

continuous mode of operation of the system will

inform the evolution of the level of collaboration both

for students about each forum that they participated

in, and for teachers about the conversations in

monitored forums.

5 CONCLUSIONS

The development described in this paper represents a

new possibility for chatbot performance to promote

an effective collaboration. The chatbot must be

implemented in an educational context more oriented

towards the construction of knowledge, which is

different from the one that is traditionally adopted.

Concisely, the research presented in this paper,

with the perspective of promoting collaborative

learning in monitored discussion forums, has

achieved the CA layer necessary to characterize the

interactions among students in the discussion forums.

In consonance with the research aims described in

section 2, the knowledge model to classify the

conversations based on the level of collaboration

among students is being developed. In the next steps,

the dialogue base will be modeled and the chatbots

will be implemented aiming to make

recommendations with suggestions and information

for students, teachers, and tutors in order to promote

collaboration among students. Moreover, chatbots

will also inform educators about the

recommendations that resulted in participation so the

constant evaluation of the ET adoption is enabled.

The evaluation of the results will occur through

the application of questionnaires to students and

teachers who commit themselves to voluntarily using

this application of AIEd, so they can assess how much

the recommendations made by the chatbots

contributed to collaborative learning.

REFERENCES

Abraham, J., Kannampallil, T., Brenner, C., Lopez, K.,

Almoosa, K., Patel, B., & Patel, V. (2016).

Characterizing the structure and content of nurse

handoffs: A Sequential Conversational Analysis

approach. Journal of Biomedical Informatics, 59, 76–

88. https://doi.org/10.1016/j.jbi.2015.11.009.

Aggarwal, C. (2016). An Introduction to Recommender

Systems. In C. Aggarwal (Ed.), Recommender Systems:

The Textbook (pp. 1–28). Springer International

Publishing. https://doi.org/10.1007/978-3-319-29659-

3_1.

Allaymoun, M. H., & Trausan-Matu, S. (2015). Rhetorical

structure analysis for assessing collaborative processes

in CSCL. 2015 19th International Conference on

System Theory, Control and Computing (ICSTCC),

123–127. https://doi.org/10.1109/ICSTCC.2015.7321

280.

Cavalcanti, A. P., Mello, R. F. L. de, Miranda, P. B. C. de,

& Freitas, F. L. G. de. (2020). Análise Automática de

Feedback em Ambientes de Aprendizagem Online.

Anais do Simpósio Brasileiro de Informática na

Educação, 892–901. https://doi.org/10.5753/cbie.sbi

e.2020.892.

Dascalu, M., McNamara, D. S., Trausan-Matu, S., & Allen,

L. K. (2018). Cohesion network analysis of CSCL

participation. Behavior Research Methods, 50(2), 604–

619. https://doi.org/10.3758/s13428-017-0888-4.

Farrow, E., Moore, J., & Gašević, D. (2019). Analysing

discussion forum data: A replication study avoiding

data contamination. Proceedings of the 9th

International Conference on Learning Analytics &

Knowledge, 170–179. https://doi.org/10.1145/33037

72.3303779.

Hayashi, Y. (2019). Multiple pedagogical conversational

agents to support learner-learner collaborative learning:

Effects of splitting suggestion types. Cognitive Systems

Research, 54, 246–257. https://doi.org/10.1016/j.cog

sys.2018.04.005.

CSEDU 2022 - 14th International Conference on Computer Supported Education

202

Koschmann, T. (2013). Conversation Analysis and

Collaborative Learning. In C. Hmelo-Silver, C. Chinn,

C. Chan, & A. O’Donnell (Eds.), The International

Handbook of Collaborative Learning (pp. 149–167).

Routledge Handbooks.

Liu, Q., Huang, J., Wu, L., Zhu, K., & Ba, S. (2019). CBET:

Design and evaluation of a domain-specific chatbot for

mobile learning. Universal Access in the Information

Society. https://doi.org/10.1007/s10209-019-00666-x.

Lu, J., Chiu, M. M., & Law, N. W. (2011). Collaborative

argumentation and justifications: A statistical discourse

analysis of online discussions. Computers in Human

Behavior, 27(2), 946–955. https://doi.org/10.1016/

j.chb.2010.11.021.

Lyndall, C.-S., & Elspeth, M. (2015). Measuring

engagement in online collaborative learning activities:

A comparative analysis of the Conversational

Framework and the Social Network Analysis tool. 2015

IEEE 3rd International Conference on MOOCs,

Innovation and Technology in Education (MITE), 100–

103.https://doi.org/10.1109/MITE.2015.7375297.

Michaels, S., & O’Connor, C. (2015). Conceptualizing Talk

Moves as Tools: Professional Development

Approaches for Academically Productive Discussions.

In L. B. Resnick, C. S. C. Asterhan, & S. N. Clarke

(Eds.), Socializing Intelligence Through Academic Talk

and Dialogue (pp. 347–361). American Educational

Research Association. https://doi.org/10.3102/978-0-

935302-43-1_27.

Oliveira, I., Tinoca, L., & Pereira, A. (2011). Online group

work patterns: How to promote a successful

collaboration. Computers & Education, 57(1), 1348–

1357. https://doi.org/10.1016/j.compedu.2011.01.017.

Pereira, G. de M. (2018). Características inerentes a

medidas de centralidade e uso de algoritmos de

aprendizado de máquina para classificação de

bridging nodes. https://doi.org/10.14393/ufu.di.2018.2

87.

Procter, M., Lin, F., & Heller, B. (2018). Intelligent

intervention by conversational agent through chatlog

analysis. Smart Learning Environments, 5(1), 30.

https://doi.org/10.1186/s40561-018-0079-5.

Saqr, M., Nouri, J., & Jormanainen, I. (2019). A Learning

Analytics Study of the Effect of Group Size on Social

Dynamics and Performance in Online Collaborative

Learning. In M. Scheffel, J. Broisin, V. Pammer-

Schindler, A. Ioannou, & J. Schneider (Eds.),

Transforming Learning with Meaningful Technologies

(pp. 466–479). Springer International Publishing.

https://doi.org/10.1007/978-3-030-29736-7_35.

Schulte, L. (2021). Investigating topic modeling techniques

for historical feature location. http://urn.kb.se/

resolve?urn=urn:nbn:se:kau:diva-85379.

Stahl, G. (2012). A view of Computer-Supported

Collaborative Learning research and its Lessons for

Future-Generation Collaboration Systems.

http://gerrystahl.net/pub/fgcs.pdf.

Stahl, G., Koschmann, T., & Suthers, D. D. (2005).

Computer-Supported Collaborative Learning. In R. K.

Sawyer (Ed.), The Cambridge Handbook of the

Learning Sciences (pp. 409–426). Cambridge

University Press. https://doi.org/10.1017/CBO9780

511816833.025.

Tawfik, A., Graesser, A., & Love, J. (2020). Supporting

Project-Based Learning Through the Virtual Internship

Author (VIA). Technology, Knowledge and Learning,

25(2), 433–442. https://doi.org/10.1007/s10758-018-

9392-x.

Tegos, S., Demetriadis, S., Psathas, G., & Tsiatsos, T.

(2020). A Configurable Agent to Advance Peers’

Productive Dialogue in MOOCs. In A. Følstad, T.

Araujo, S. Papadopoulos, E. L.-C. Law, O.-C. Granmo,

E. Luger, & P. B. Brandtzaeg (Eds.), Chatbot Research

and Design (pp. 245–259). Springer International

Publishing. https://doi.org/10.1007/978-3-030-39540-

7_17.

Uthus, D. C., & Aha, D. W. (2013). Multiparticipant chat

analysis: A survey. Artificial Intelligence, 199–200,

106–121. https://doi.org/10.1016/j.artint.2013.02.004.

Vayansky, I., & Kumar, S. A. P. (2020). A review of topic

modeling methods. Information Systems, 94, 101582.

https://doi.org/10.1016/j.is.2020.101582.

Zawacki-Richter, O., Marín, V. I., Bond, M., &

Gouverneur, F. (2019). Systematic review of research

on artificial intelligence applications in higher

education – where are the educators? International

Journal of Educational Technology in Higher

Education, 16(1), 39. https://doi.org/10.1186/s41239-

019-0171-0.

Conversational Analysis to Recommend Collaborative Learning in Distance Education

203