Integrating Automated and Humanistic Approaches: A

Methodological Case Study of Teachers' Digital Professional Growth

Corrado Matta

, Susanna Nordmark

, Kristina Holmberg

, John Rack

Mattias Davidsson

and Italo Masiello

Department of Education, Linnaeus University, Universitetsplatsen 1, Växjö, Sweden

Department of Computer Science and Media Technology, Linnaeus University, Universitetsplatsen 1, Växjö, Sweden

Keywords: Methodological Integration, Digital Learning Materials, Text Mining, Discourse Analysis, AI-Supported

Qualitative Analysis, Large Language Models.

Abstract: This paper presents a methodological case study on teachers' digital professional development, emphasizing

the integration of automated and humanistic approaches. Drawing from a four-year pilot project led by the

research group, we explore how three distinct analytical methodologies—manual discourse analysis, text

mining, and large language model-assisted thematic analysis—were employed to examine teachers' discursive

practices regarding digital learning materials. The study investigates how integrating these methodologies

enhances our understanding of digital learning material-related discourses and their evolution over time. Key

findings reveal two primary conceptualizations: digital learning materials as pedagogical/effectivization tools

and as complementary to analogue resources. The integrated approach demonstrated advantages in mitigating

methodological biases, improving reliability, and enabling a richer analysis of diverse data sources. This work

contributes to the development of robust analytical frameworks for studying the intersection of technology

and pedagogy in educational settings.

1 INTRODUCTION

In this paper, we describe and discuss a case study of

methodological integration applied to a study of

teachers' digital professional development. The paper

aims (a) to describe how three different analytical

approaches – large language model (LLM), text

mining, and traditional humanistic discourse analysis

– were integrated to study the discursive dimension

of teachers' digital professional development and (b)

to discuss the methodological advantages of this

integration.

The paper begins in Section 2 with a description

of the background project of which the described

experiment is a part. Section 3 describes the three

different methodologies used in our case and their

results. In section 4, we explain how the results from

https://orcid.org/0000-0003-2282-8071

https://orcid.org/0000-0001-7313-1720

https://orcid.org/0000-0002-2924-4100

https://orcid.org/0000-0001-7525-6180

https://orcid.org/0000-0002-9474-6879

https://orcid.org/0000-0002-3738-7945

the three approaches were integrated and discuss the

advantages of our integrative approach.

2 BACKGROUND

Guided by principles of Implementation Science, the

research group at Linnaeus University has led a four-

year pilot project to enhance teachers’ digital

competencies, foster data-driven learning, and utilize

Visual Learning Analytics (Masiello et al., 2023;

Nordmark et al., 2024). This collaboration included

researchers, municipalities, schools, and leading

providers of Digital Learning Materials (DLMs) in

Sweden.

The project has involved extensive data

collection, including interviews, observations,

458

Matta, C., Nordmark, S., Holmberg, K., Rack, J., Davidsson, M. and Masiello, I.

Integrating Automated and Humanistic Approaches: A Methodological Case Study of Teachers’ Digital Professional Growth.

DOI: 10.5220/0013488100003932

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Conference on Computer Supported Education (CSEDU 2025) - Volume 2, pages 458-464

ISBN: 978-989-758-746-7; ISSN: 2184-5026

logbooks, and surveys. As the data analysis phase

started, the team agreed on a set of sub-studies, each

providing a part of the picture of teachers' digital

professional development.

The first sub-study the team decided to work on is

the object of this paper. This first sub-study

investigates the participating teachers' discursive

practices centered on DLMs and was motivated by

the following research questions:

RQ1: What ideas about the concept and role of

DLMs emerge from the perspectives of actors

involved in a school digital transformation

project? How do these ideas vary over time and

across different participants?

RQ2: What expectations regarding DLMs are

reflected in the data? How do these expectations

differ over time and among various participants?

2.1 Methodology of the Sub-Study

The sub-study focused only on the teacher and school

principal interview data. This data set consisted of

transcripts from 22 semi-structured interviews with

44 teachers and three school principals. The

interviews were organized into three rounds over

three years, one round each year.

All interviews were transcribed, segmented (one

segment for each turn-taking), and diarized. The data

set was annotated specifying, for each segment:

speaker, school, municipality, taught subjects, grade,

date of the interview, interview round, and the DLM

used.

After discussing which analytical approach to

use, the team determined to employ three different

approaches. Each approach and its results are

described in the next section.

3 ANALYSES AND RESULTS

This section contains a discussion of the three

analytical approaches used in three sub-studies of

Teachers' DLM discourses. The three different lines

of analysis are discussed in detail in the three separate

studies (Holmberg et al., 2025; Masiello et al., 2025;

Matta et al., 2025), while this article collects the

aggregate summary analysis. Each of the following

three subsections summarizes the methodology and

results of one of these three studies.

The summaries in this section are brief and only

describe the methodology and the most central results

of the separate studies. We discuss the validity of

each analytical approach in section 4.2 and refer the

reader to the individual studies for a discussion of the

reliability of the instruments.

3.1 First Approach: Manual Discourse

Analysis

This section presents the discourse analysis of the

interview data (Holmberg et al., 2025), guided by

Laclau and Mouffe's (1985) framework and Gee’s

procedural approach (Gee, 2001), which emphasize

agency and the creation of discursive patterns within

social practices. We call this approach humanistic to

emphasize the centrality of the researchers'

interpretive competence as the main analysis tool

(Pääkkönen & Ylikoski, 2021). Discourses are

conceived as semiotic dimensions of social practices

(in this case: the use of DLMs). Through participation

in the project, teachers and principals develop

linguistic repertoires that shape their representations

of what DLMs are, their functions, and their ideal

characteristics. These representations, termed

discursive formations, are inferred through

observable linguistic activities.

Two members of the research team worked on the

analysis of the data, which involved three steps. First,

the data were reduced by identifying all the segments

related to DLM concepts and functions. Secondly, the

reduced data were coded by grouping relevant

segments thematically to reflect participants'

conceptualizations, idealizations, and expectations of

DLMs. Finally, discursive formations were identified

by classifying themes into distinct discourses

representing conceptualizations, idealizations of,

and/or expectations about DLMs.

3.1.1 Results and Interpretation

The list of themes resulting from the coding process

and their classification into discursive formations is

found in the Appendix. The analysis resulted in three

representations of the concept and idealized image of

a DLM.

The first conceptualization conceives the DLM as

a Pedagogical Tool and focuses on DLMs as

instruments to enhance learning outcomes,

emphasizing DLMs as capable of improving teacher-

student interaction, generating multimodal

representation of subject contents, supporting and

scaffolding reasoning, and affecting learning goals.

Hence, DLMs have an active impact on students'

learning.

In contrast, the second conceptualization depicts

the DLM as an Effectivization Tool. According to this

Integrating Automated and Humanistic Approaches: A Methodological Case Study of Teachers’ Digital Professional Growth

459

idealization, DLMs are a means to streamline

educational processes, enabling task optimization,

monitoring, communication with students or parents,

and integration while maintaining teacher control.

This conceptualization differs from the first, as DLMs

have a more infrastructural and peripheral role in the

teacher's pedagogy. They do not impact learning

directly but support the teacher in all the activities that

are ultimately aimed at students' learning.

The final conceptualization focuses on DLM and

the Digital/Analog Divide. This final

conceptualization highlights tensions between digital

and analog materials, with "hybrid" uses of DLMs

conceptualized as a balance, especially towards the

end of the project.

The findings show that the teachers' discursive

practices changed over time. Participants gained

deeper insights into the strengths and limitations of

these tools, improving their ability to integrate them

effectively. Moreover, changes in the teachers'

conceptualizations might also be a result of a shift in

the broader social discourse concerning the

digitalization of the educational sector. This

interpretation concerns the third discursive formation,

which emerged towards the end of the project parallel

to an ongoing shift in the public discussion about the

use of digital tools in schools.

3.2 Second Approach: Text Mining

The second approach employed a text-mining

analysis of the interview data (Matta et al., 2025),

using statistical algorithms applied in Python (version

3.12.3) and R (version 2024.09.1+394).

A text corpus was built using the interview data

sets, where each segment constituted a document. The

annotated information was included in the corpus as

document variables. The resulting corpus consisted of

2262 documents, 8 variables, and 34730 tokens (i.e.,

the number of words in the whole corpus), with a

mean of 15.36 tokens per document and a standard

deviation of 17.64 tokens.

Four analytical tools were used to analyze the

corpus.

First, we performed sentiment analysis using the

KBLab Sentiment Analysis classifier, a transformer-

based neural network developed by KBLab at the

Swedish Royal Library (Hägglöf, 2023). Trained on

a dataset of 165,000 manually labeled Swedish texts.

The classifier categorizes sentiment into positive,

neutral, or negative with an estimated accuracy of

80%.

The second approach we employed was

Correspondence Analysis (FactoMineR, Factoextra).

This approach explores associations between

sentiment – i.e., values of the categorical variable

"sentiment" – and project rounds. Using dimension

reduction (Singular Value Decomposition), it mapped

the co-occurrence of categorical values in a two-

dimensional space (Husson et al., 2024; Kassambara

& Mundt, 2020).

The third analysis we applied was Keyness

Analysis (Quanteda Textstats). Here, we identified

words and phrases significantly differing across

rounds. Terms from the final round were compared

with earlier ones using Chi-squared statistics (Benoit

et al., 2024).

Finally, we used Topic Modeling (SeededLDA),

and, more specifically, Applied Latent Dirichlet

Allocation (LDA) to uncover thematic clusters in the

data, identifying topics as word groups with shared

themes (Watanabe & Xuan-Hieu, 2024).

3.2.1 Results and Interpretation

Correspondence Analysis: Sentiment evolved from

neutral in the early stages to negative mid-project and

positive near the end. This reflected initial technical

concerns, giving way to constructive views as

participants adapted to DLMs.

Keyness Analysis: Early discussions focused on

technical issues (“computer,” “platform”), while later

rounds emphasized pedagogical aspects

(“understand,” “exercise”) and the complementary

role of DLMs alongside traditional materials

(“complement,” “book”). Participants increasingly

framed DLMs as supplementary resources rather than

replacements.

Topic Modeling: Identified five relevant topics,

forming two main thematic clusters: (1) DLMs as

pedagogical/effectivization tools and (2) DLMs in

relation to analog materials.

Keyness analysis and topic modeling revealed two

primary representations of DLMs: DLMs as

pedagogical/effectivization tools and as

complementary to analog teaching materials. The first

representation emphasizes the use of DLMs either as

tools that impact learning directly or as tools that

simplify teachers' work. The choice of merging these

two conceptions into a single representation and not

as two different ideas as in the case of manual

discourse analysis was mainly based on topic

modeling, which indicated that the pedagogical and

effectivization features often occurred together.

The second representation, more clearly emerging

as an independent thematic cluster, emphasized the

CSEDU 2025 - 17th International Conference on Computer Supported Education

460

tension between digital and analog pedagogical tools,

where DLMs were discussed only in relation to this

tension. We introduced here the concept of boundary

object (Fleischmann, 2006; Fox, 2011) to describe the

DLMs. According to this theory, the way of using

language to conceptualize and idealize DLMs

emphasizes the ongoing debate in the educational

sector concerning the primacy of digital or analog

pedagogical approaches and sees the DLM as an

object that constitutes and maintains this tension.

Correspondence analysis indicated that sentiment

shifted towards positive over time, which is

consistent with the interpretation of the data in the

manual discourse analysis, indicating that teachers

developed a deeper understanding of educational

technologies.

3.3 Third Approach: LLMs

The third analytical approach was to employ LLMs to

support the thematic analysis of the interview data

(Masiello et al., 2025).

The analysis was conducted using the ChatGPT-

4o, guided by Braun and Clarke's (2013) thematic

analysis framework. This third analysis focused on

teachers' and school principals' systems of

expectations about DLMs. The analysis started with

an initial coding, where relevant data segments were

identified based on recurring topics and phrases, and

a custom stop-word list was applied during

preprocessing. To assess thematic relevance,

recurring terms (e.g., we have, has been, not really)

were extracted. The second step involved developing

themes from the initial codes. Codes were grouped

into themes such as Teacher Confidence, DLM

Integration, Student Engagement, Challenges, and

Outcomes. Themes were refined iteratively for

accuracy. The analysis concluded by focusing on

temporal comparisons. Themes were analyzed across

different stages (e.g., early vs. late project phases) to

identify shifts in perspectives.

3.3.1 Results and Interpretation

The analysis highlighted key themes, categories, and

codes:

Expectations and Idealized Views

 Interactive and Dynamic Learning:

Anticipated enhanced student engagement

through features like multimedia,

interactivity, and gamified learning (fun,

interactive, video, and games).

 Personalized Learning: Belief in digital

tools’ ability to tailor lessons to individual

needs (adjust content, tailor lessons,

review).

 Teaching Efficiency: Expected to simplify

workloads with automated grading and

resource organization (automate tasks, save

time).

Challenges with Content and Implementation

 Content Quality and Alignment: Digital

materials often lacked depth or curriculum

alignment (not aligned, fit to lesson).

 Technical and Logistical Barriers: Teachers

faced platform difficulties, glitches, and

steep learning curves (troubleshooting, not

simple).

 Student Engagement and Digital Literacy:

Not all students adapted well to digital tools,

with varying competence levels (not

comfortable, not engaging).

Temporal Evolution of Perspectives

 Early Phase (2021): Optimism about digital

tools’ potential for innovation, engagement,

and personalization (transform teaching,

make it easier).

 Later Phase (2024): A pragmatic focus on

high-quality content and effective

integration, with less emphasis on

transformative change (takes time, need to

adapt).

This analysis revealed how expectations of DLMs

evolved over time, providing valuable insights into

their adoption and integration into educational

practices.

4 METHODOLOGICAL

DISCUSSION

This section describes the approach used to integrate

and compare the three different lines of analysis and

discusses the advantages and limitations of this

approach.

4.1 Integrative Procedure

After having decided to approach our research

questions from three different analytical perspectives,

the team agreed on using an iterative and explorative

Integrating Automated and Humanistic Approaches: A Methodological Case Study of Teachers’ Digital Professional Growth

461

procedure as a methodological approach for

comparing and integrating the different insights.

The first stage of the procedure, after having

determined the research questions, was to form three

teams (each working with one of the analytical

approaches), and each team would generate

preliminary insights on their own. Next, a first

comparison meeting was arranged in which each team

presented their preliminary insights. Then, it was

decided to interpret each insight not as a result that

could be interpreted as an answer – albeit tentative –

to our research questions, but rather as a new point of

departure, an avenue for further analysis. This

entailed ascribing limited credibility to the

preliminary results and going forward with the three

separate lines of analysis. Finally, a final meeting was

arranged to compare the outcomes of the different

analyses.

The results described in Section 3 represent the

outcomes of this iterative and exploratory approach.

The comparison of these outcomes revealed that the

three lines of analysis converged on two discursive

formations: the pedagogical/effectivization tool

discourse and the analog/digital discourse. The next

section discusses what level of credibility can be

ascribed to the claim that these discourses indeed

represent the structure of language use among the

participants.

4.2 Advantages of Our Approach

The issue of the credibility of an interpretive theory,

such as that which was generated by our

iterative/explorative approach, is, in essence, a matter

of distinguishing a plausible interpretation from an

interpretive artifact. An interpretation is credible if it

is likely to represent the actual social phenomena it

targets. According to an inferentialist/pragmatist

perspective (Suárez, 2004), which we assume,

representation allows agents to make inferences about

its target phenomenon. This means that an

interpretation is credible if it can be used to make

fruitful explanations and projections about the target.

In contrast, an interpretive artifact is simply a result

of forcing a narrative onto the data, which typically

results in unreliable inferences.

Assessing the credibility of an interpretation is

best done by assessing the risks of interpretive

For instance, someone could interpret a chair as a jacket.

This interpretation will allow the interpreter to infer that

wearing it will warm her/him. The interpretation is made

less credible by the fact that it generates unreliable

inferences.

artifacts and discussing the methodological ways to

manage these risks. Three types of methodological

risks affect our case, one for each of the lines of

analysis.

Manual analysis relies on the interpretive

schemes of the researcher and is, therefore, easily

affected by individual biases. Humanistic

interpretation rests on selecting what interpretations

seem to best fit the data (Matta, 2022). Individual

selection criteria can be biased towards explanatory

schemes that are familiar or otherwise preferred,

which introduces implicit weights or bias in the

selection. As the process advances in the ladder of

interpretation (coding → thematization →

interpretation in terms of discursive formations), the

risk of a researcher projecting such preconceptions on

the data increases.

Text mining is based on statistical algorithms,

which involve a risk of generating statistical artifacts.

Researchers might be tempted to interpret statistical

patterns as meaningful insofar as they are statistical

patterns, but this increases the risk that some of the

observed patterns in the data are simply statistical

artifacts – that is patterns detected by the algorithm

that depend systematic errors in the analysis or the

data collection – and not existing relationships in the

target phenomenon. For instance, the LDA algorithm

used in our study can sometimes generate spurious

topics by clustering terms that are lexically identical

but used in different ways in different contexts.

Finally, using LLMs in research can be affected

by different types of biases. Several sources (Ashwin

et al., 2023; Schroeder et al., 2024) have discussed

how using LLM for qualitative analysis might

increase such risk. One source of bias can be the

natural language data that is used to train the LLMs,

which can be representative of other social groups

than that which is analyzed. This will increase the

probability that the chosen LLM produces

interpretations that fit the context of the training data.

Another source of bias can originate from the

concepts and theories used to train the LLMs. It is

important to highlight that LLMs do not generate

analyses but report a statistical synthesis of the

formulations used in training texts that the algorithm

categorizes as analyses. Hence, whenever a LLM

proposes a thematization, it is not proposing a model

of the data but rather trying to summarize the textual

CSEDU 2025 - 17th International Conference on Computer Supported Education

462

behavior of interpretations included in the LLMs'

training data. As a result, LLMs will more likely

"interpret" the interview data according to more

recurring interpretive frameworks, which introduces

a conservative bias in the interpretation (bolder

interpretations are systematically excluded).

Our approach has several rewards, some of which

provide strategies that manage – although do not

eliminate – all of these risks. First, it exploits the

advantages of automatized and humanistic

approaches by combining the depth of humanistic

interpretation with the breadth of automatized

procedures. Manual interpretations are more

sophisticated but cannot manage large datasets,

whereas automatized methods allow for the analysis

of large datasets but are typically superficial. Our

approach establishes a balance between these two

dimensions.

Secondly, its explorative and iterative character

contributes to the outcome's reliability. Ascribing a

lower level of credibility to the preliminary insights

decreases the risk of falling for compelling narratives.

Moreover, it harnesses the value of

methodological pluralism as an analytical tool by

working with three separate lines of analysis and

comparing iteratively the insights of all the

approaches; there is no single method of analysis that

acquires a leading position. This avoids the typical

bias toward quantitative analyses, which affects many

mixed-methods studies. This bias is the result of a

methodological assumption, according to which

qualitative methods are appropriate for hypothesis

generation and quantitative methods are best for

theory testing. We challenge this view by focusing on

how the target phenomenon was modeled using the

different approaches and reflecting on the

assumptions that these models inherit from each

approach.

Our approach provides a management strategy for

the bias risks mentioned above. The iterative and

comparative approach decreases the risk for both

individual, statistical, and LLM-based biases by

letting each line of analysis work as a watchdog for

all others. If the LLM generates an analysis that is

biased toward a social group, the humanistic analysis

is likely to pick up that bias in virtue of its sensitivity

for context. In the same way, if the humanistic

interpretation is biased by the researcher's interpretive

scheme, there is a chance that both the LLM and the

text mining analyses will fail to confirm that insight,

as the latter are less prone to cherry-picking.

Furthermore, if the text mining analysis is based on a

statistical artifact, the humanistic interpretation will

plausibly find that pattern far-fetched by identifying

spurious topics. Finally, the inclusion of a humanistic

component in the analysis allows for bolder

interpretation, decreasing the risk of interpretive

conservatism.

It is important to highlight that our integrative

approach provides strategies that manage epistemic

risks but cannot eliminate these risks. We cannot

exclude the possibility that the individual researchers

who apply the humanistic interpretive approach will

not suffer the same kind of biases that could affect the

LLMs and text mining approaches. A human

researcher can suffer from a conservative bias or miss

a spurious topic by failing to catch that the same word

is used in different ways throughout a data set.

However, by pluralizing the analysis of qualitative

data, the risk of such biases is arguably lower than

when working with any of the three approaches in our

substudy.

5 CONCLUSIONS

The integration of manual, automated, and LLM-

assisted methodologies in this study has highlighted

the value of methodological pluralism in educational

research. By combining humanistic insights with

automated and AI-supported approaches, we were

able to uncover teachers' perspectives on the evolving

roles of DLMs in school. This iterative and

exploratory approach not only mitigated biases

inherent in individual methodologies but also helped

in generating a comprehensive understanding of

DLM discourses.

The findings emphasize the potential of DLMs as

tools for both pedagogical enhancement and

operational efficiency while also revealing ongoing

tensions between digital and analog educational

resources. Our study underscores the importance of

an interdisciplinary lens in addressing complex

educational challenges, offering a methodological

approach for integrating diverse analytical

perspectives. Future research could expand this

framework to other educational contexts, further

validating its applicability and effectiveness.

ACKNOWLEDGMENTS

An initial sketch of the abstract and conclusion of this

paper was generated using an AI tool (ChatGPT-4o).

The sketch was substantially revised, and the present

versions of the abstract and conclusion retained less

than 40% of these initial sketches.

Integrating Automated and Humanistic Approaches: A Methodological Case Study of Teachers’ Digital Professional Growth

463

REFERENCES

Ashwin, J., Chhabra, A., & Rao, V. (2023). Using Large

Language Models for Qualitative Analysis can

Introduce Serious Bias (arXiv:2309.17147). arXiv.

https://doi.org/10.48550/arXiv.2309.17147

Benoit, K., Watanabe, K., Wang, H., Lua, J. W., Kuha, J.,

& Council (ERC-2011-StG 283794-QUANTESS), E.

R. (2024). quanteda.textstats: Textual Statistics for the

Quantitative Analysis of Textual Data (Version 0.97.2)

[Computer software]. https://cran.r-

project.org/web/packages/quanteda.textstats/index.htm

Braun, V., & Clarke, V. (2013). Successful qualitative

research: A practical guide for beginners (1. ed.).

SAGE Publications.

Fleischmann, K. R. (2006). Boundary Objects with

Agency: A Method for Studying the Design–Use

Interface. The Information Society, 22(2), 77–87.

https://doi.org/10.1080/01972240600567188

Fox, N. J. (2011). Boundary Objects, Social Meanings and

the Success of New Technologies. Sociology, 45(1),

70–85. https://doi.org/10.1177/0038038510387196

Gee, J. P. (2001). An introduction to discourse analysis

[Elektronisk resurs] theory and method. Routledge.

Hägglöf, H. (2023). The KBLab Blog: A robust, multi-label

sentiment classifier for Swedish. https://kb-

labb.github.io/posts/2023-06-16-a-robust-multi-label-

sentiment-classifier-for-swedish/

Holmberg, K., Matta, C., Nordmark, S., Masiello, I., Rack,

J., & Davidsson, M. (2025). From Balance to Informed

Criticism – Discursive Transformations in a

Development Project on Digital Learning Material in

Primary School. Proceedings of INTED Conference.

Husson, F., Josse, J., Le, S., & Mazet, J. (2024).

FactoMineR: Multivariate Exploratory Data Analysis

and Data Mining (Version 2.11) [Computer software].

https://cran.r-

project.org/web/packages/FactoMineR/index.html

Kassambara, A., & Mundt, F. (2020). factoextra: Extract

and Visualize the Results of Multivariate Data Analyses

(Version 1.0.7) [Computer software]. https://cran.r-

project.org/web/packages/factoextra/index.html

Laclau, E., & Mouffe, C. (1985). Hegemony & socialist

strategy. Verso.

Masiello, I., Fixsen, D. L., Nordmark, S., Mohseni, Z.

(Artemis), Holmberg, K., Rack, J., Davidsson, M.,

Andersson-Gidlund, T., & Augustsson, H. (2023).

Digital transformation in schools of two southern

regions of Sweden through implementation-informed

approach: A mixed-methods study protocol. PLOS

ONE, 18(12), e0296000.

https://doi.org/10.1371/journal.pone.0296000

Masiello, I., Matta, C., Holmberg, K., Nordmark, S., Rack,

J., & Mohseni, Z. (2025). An AI Chat-Bot Thematic

Analysis of Teachers’ Expectations of Digital Learning

Materials in Primary Schools. Proceedings of INTED

Conference.

Matta, C. (2022). Tolkningens metodologi och selektiv

abduktion. Pedagogisk forskning i Sverige,

27(4), 37–

61. https://doi.org/10.15626/pfs27.04.03

Matta, C., Nordmark, S., Holmberg, K., Davidsson, M.,

Rack, J., & Masiello, I. (2025). The Discursive

Mechanisms of Teachers’ Digital Professional

Development: A Quantitative Approach. Proceedings

of INTED Conference.

Nordmark, S., Augustsson, H., Davidsson, M., Andersson-

Gidlund, T., Holmberg, K., Mohseni, Z. (Artemis),

Rack, J., & Masiello, I. (2024). Piloting Systematic

Implementation of Educational Technology in Swedish

K-12 Schools – Two-Years-In Report. Global

Implementation Research and Applications, 4(3), 309–

323. https://doi.org/10.1007/s43477-024-00130-w

Pääkkönen, J., & Ylikoski, P. (2021). Humanistic

interpretation and machine learning. Synthese, 199(1),

1461–1497. https://doi.org/10.1007/s11229-020-

02806-w

Schroeder, H., Quéré, M. A. L., Randazzo, C., Mimno, D.,

& Schoenebeck, S. (2024). Large Language Models in

Qualitative Research: Can We Do the Data Justice?

(arXiv:2410.07362). arXiv.

https://doi.org/10.48550/arXiv.2410.07362

Suárez, M. (2004). An Inferential Conception of Scientific

Representation. Philosophy of Science, 71(5), 767–779.

https://doi.org/10.1086/421415

Watanabe, K., & Xuan-Hieu, P. (2024). seededlda: Seeded

Sequential LDA for Topic Modeling (Version 1.4.1)

[Computer software]. https://cran.r-

project.org/web/packages/seededlda/index.html

APPENDIX

Discourse Themes

DLM as a

Pedagogical Tool

Teacher-adapted,

Language functions,

Multimodality, Learning

analytics,

Fun/Lively/Interesting,

Students' Digital

Competence, Reasoning,

Flexibility, Explanations,

Goal Fulfillment,

Repetition, La

out.

DLM as an

Effectivization Tool

Efficiency, Assessment,

Specific Applications,

Integration, Monitoring,

Family, Language

functions, Repetition,

Teacher-adapted,

Updated.

DLM and the

Digital/Analog

Divide

The divide between

digital and analog

materials, h

brid solutions

CSEDU 2025 - 17th International Conference on Computer Supported Education

464