At ECER 2024 the presentations of two projects on identifying and analysing thematic trends and clusters of the contributions to ECER since 1998 gained much attention and resulted in interesting and fruitful discussions. Both presentations utilized topic modelling (Blei et al., 2003; Griffiths & Steyvers, 2004) to determine the central content-related themes of over 30,000 contributions ranging over 25 years. At its core, topic modelling is a natural language processing and machine learning method clustering documents in regard to their thematic similarity based on the words occurring in the documents frequently (together).

While both projects utilized topic modelling for analysing the same corpus, they differed in their central goals and approaches to interpreting the results: The presentation by Hoveid et al. (2024) utilized a more qualitative approach to analyse and interpret the results of a model consisting of k = 100 thematic topics to determine, assess and compare thematic trends of ECER since 1998. The second project by Schindler et al. (2024) was based on the development of the interactive web app EduTopics: ECER (Christ et al., 2024; url: https://dipf-lis.shinyapps.io/EduTopicsECER/) which enables users to explore clusters and trends within the corpus. This was made possible by providing univariate and multivariate interactive and manipulatable visualizations for bibliographic parameters of the contributions (EERA network, authors, country of affiliation) and their relevance for the k = 50 thematic clusters resulting from the topic modelling approach. During the research workshop on the app, examples of applications from various (sub-)disciplinary or EERA network perspectives were presented and discussed. The following discussion contained several relevant aspects for further developments of the app’s functionalities or application areas.

This research workshop addresses both aforementioned aspects by including presentations on (a) new developments regarding the functions and visualisations and (b) highlighting use-cases of the app from various different EERA network perspectives.

The first presentation is on new developments regarding the app including an update to the corpus with the ECER 2025 contributions. Updates to existing functions as well as new functions are presented, especially the new and improved hierarchical topic modelling approach, which addresses the main point of contention during the discussion in 2024 on the feasibility of clustering over 30,000 documents in only k = 50 topics (see methods and conclusion sections).

Another presentation is on the participation at ECER conferences across the years. This presentation aims at the identification of predictors for explaining the variance within the contribution numbers of affiliation countries across the years.

Two additional presentations are based on the application of EduTopics for research use-cases:

The first use-case focuses on clusters, trends and central nodes of research on teacher education at ECER both within network 10 “Teacher Education Research” and beyond.

The second uses EduTopics to develop an overview of trends and clusters of contributions from network 22 “Research in Higher Education”.

The final presentation gives a brief look at the future of EduTopics including the integration of additional data sources for comparative analyses of different conferences and national databases on educational research.

Finally, ample time for discussions and feedback on the presentations and the app is allocated to gather further information for additional (technical) developments, use-cases and applications of the app.    

Presentation EduTopics: ECER 2025 Addressing the issue on the fit of a small number of topics and a large number of contributions, a hierarchical, nested topic modelling approach is utilized by simultaneously computing (1) a model with a large number of topics (between k = 200 and k = 300) capturing smaller thematic clusters giving the heterogeneity of educational research justice and (2) a broader model with a smaller number of topics (around k = 50), which enables an efficient overview over the central themes within the corpus. The similarities between the topics of both models are determined by aggregating various distance and similarity measures both on the document and word level (Aletras & Stevenson, 2014; Blair & Mulvenna, 2020). The aggregated similarity measure is utilized to determine the nesting of the subtopics (of model 1) within the supertopics (model 2). This hierarchical representation of the themes is utilized to determine which smaller, more detailed topics are most similar to the larger, broader topics. This enables users to gain insight into broad, overarching themes and their more fine-grained sub-topics such as varying trends of the topics “special needs education” and “inclusive education” in the overarching topic “special needs and inclusive education”. Presentation Country-Level-Data: Potential predictors, like size of national higher education systems or distance to conference venue, for the longitudinal and cross-sectional variance within the number of contributions by country are identified, presented and tested regarding their incremental explanation value within a multi-level regression model. The country-level regression intercepts – i.e. the controlled contribution rate – and the country-level longitudinal slope factors are then used to identify factors leading to in- and decreases of contributions by countries over the years. Presentations Use-Cases EduTopics: Both presentations will use the available functions of the app to determine clusters, trends and potential desiderata within their respective foci on networks and themes. Those include univariate trend analyses of contributions by EERA networks and trends of the thematic topics, relationships between networks and topics via the graph visualisations and relationships between the overarching themes (from model 1) and the detailed themes (from model 2).

The presentations in this workshop give insight into the current and future development of the app EduTopics and use-cases of the app specifically but also of natural language processing and machine learning in general for educational research purposes. By applying different perspectives during the different presentations on the apps’ functions, results and visualisations, exemplary use cases and avenues for future research are highlighted. In addition, the presentations on the use-cases of the app show the potential of the combination of different research approaches. On the one side, the app opens up new ways for researchers to access data, which may not have been possible before. On the other side, the presentations of the use-cases of the app provide important feedback for the development of the underlying algorithms, (visual) presentations of the results and for the user interface and its accompanying explanations. Furthermore, the discussion will focus on current and potential future challenges for modelling heterogeneous literature corpora such as data quality, multilingualism or limitations of the applied methods in itself. We welcome additional feedback from the audience during the discussion and are open to answer ad-hoc research questions posed by audience members.

Aletras, N., & Stevenson, M. (2014, April). Measuring the similarity between automatically generated topics. In Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, volume 2: Short Papers (pp. 22-27). Blair, S. J., Bi, Y., & Mulvenna, M. D. (2020). Aggregated topic models for increasing social media topic coherence. Applied Intelligence, 50, 138-156. https://doi.org/10.1007/s10489-019-01438-z Blei, D. M., Ng, A. Y., & Jordan, M. I. (2003). Latent dirichlet allocation. Journal of machine Learning research, 3(Jan), 993-1022. Christ, A., Röschlein, J., & Schindler, C. (2024). EduTopics: ECER - An interactive app for visualising and exploring the contributions of ECER-conferences for 1998 to 2024. https://dipf-lis.shinyapps.io/EduTopicsECER Griffiths, T. L., & Steyvers, M. (2004). Finding Scientific Topics. PNAS. Hoveid, M.H., Deleye, M., Ciolan, L., Rodriguez, C.C., & Figueiredo, M.P. (2024). Thematical Trends in 30 Years of European Educational Research. Looking Back to Look Ahead. ECER 2024, Cyprus. Schindler, C., Christ, A., Röschlein, J., Cronqvist, M., & Keiner, E. (2024). Thematical Trends in 30 Years of European Educational Research. Looking Back to Look Ahead. ECER 2024, Cyprus.