Masterclass in Text Analysis

Emilia Wiśnios and Bartosz Pieliński, University of Warsaw

Introduction to large language models (LLMs) implementations into NLP tasks

This workshop aims to give participants a basic but comprehensive understanding of large language models (LLMs) and demonstrate how they can implement them in their research tasks related to text analysis. To achieve this aim, several questions are attempted to be answered:

• What is a language model?
• What are LLMs?
• What are the significant characteristics of LLMs?
• What are their distinct features?
• How are LLMs evaluated?
• How to choose a LLM?

The second part of the workshop will be dedicated to practical exercises showing how to implement LLMs into text analysis using in-context learning. Participants will receive a notebook where they will use different forms of prompting to fulfil simple NLP tasks. The hope is that this experience will show potential ways of incorporating LLMs into their research.

Prerequisites:

1. Create a free account on Hugging Face:
   1. Visit huggingface.co
   2. Click on the “Sign Up” button in the top right corner of the page and follow the instructions.
2. Accept the LLaMA 3.2 License:
   1. After logging in, visit the LLaMA 3.2 model page (meta-llama/Llama-3.2-1B · Hugging Face).
   2. Click on the “Accept” button to agree to the LLaMA 3.2 licence terms. You have to expand the upper left part of the webpage.

You may need to fill in some information or confirm that you are using the model for non-commercial purposes if prompted.

Note: Temporary downtimes may occur during the exercise while models perform calculations. During these moments, we welcome discussions about applying LLMs to your research projects.

Bio: 

Bartosz Pieliński is an Assistant Professor at the Faculty of Political Science and International Studies at the University of Warsaw. He collaborates with MI²DataLab and the UNESCO Chair on Intangible Cultural Heritage, bridging social sciences and computer science. More information on him: pielinski.info

Emilia Wiśnios is an alumna of the University of Warsaw, holding a Master’s degree in Machine Learning. Ex-Samsung. Her expertise and interests are primarily focused on Natural Language Processing (NLP), with a particular emphasis on language models. More info: https://www.linkedin.com/in/emilia-wisnios/

(detailed information will be available successively)

Dariusz Galasiński, University of Wrocław

Workshop in discourse analysis within the perspective of critical discourse studies and critical linguistics

Dirk Hovy, Bocconi University

Text analysis for social science in Python

This workshop will provide an in-depth examination of the various techniques used in text analysis, an important component in today’s information-rich society. We will reroute the main components of the paper “Capturing Regional Variation with Distributed Place Representations and Geographic Retrofitting”, both in theory and practice (via Jupyter notebooks).

Participants will learn about text representations and how to use and interpret modern embedding models effectively. The course will also cover text classification techniques and clustering, which are critical for understanding complex patterns in language data. We will finally cover geocoding, which is a method for incorporating geographical context into text-based analysis.

This workshop provides a comprehensive practical guide for social scientists interested in using Python to dissect and comprehend text data.

Grzegorz Bryda, Jagiellonian University

Computer-assisted content analysis using AI and large language models

The workshop aims to introduce participants to computer-assisted analysis of textual data using artificial intelligence and large language models. It will explicitly demonstrate the capabilities of MAXQDA software. The research method will showcase the methodologies of dictionary-based content analysis, detailing the use of MAXQDA for lexical and semantic coding procedures. The analysis approach will utilise MAXQDA to analyse and visualise the relationships between codes and categories. A dictionary model of knowledge representation will be constructed within the software, emphasising hands-on experience with text analysis software and AI tools. The workshop combines qualitative and quantitative content analysis elements to thoroughly examine and interpret textual data. The workshop methodology focuses on interactive sessions during which participants use data in real-time.

Michał Krzyżanowski, Uppsala University

Critical Discourse Studies in/and the Exploration of Social Change: Analysing Discursive Shifts

This masterclass offers a crash introduction to how Critical Discourse Studies (CDS) in general – and its central, Discourse-Historical Approach (or DHA) in particular – can be applied to study dynamics of public discourse in relation to various facets of social, political and institutional change. The masterclass highlights how deploying a multilevel CDS/DHA – and utilising such of their central analytical and interpretive notions as ‘discursive strategies’, ‘recontextualization’, and in particular of ‘discursive shifts’ – can help critically and systematically analyse and deconstruct discursive dynamics as well as recontextualization of discursive strategies across traditional and online (including social) media and various modes of political, policy and institutional communication. The masterclass will be divided into three parts. The first part will deal with a general introduction to Critical Discourse Analysis by focussing on its key assumptions, its evolution into wider Critical Discourse Studies and while offering a quick overview of the key approaches in contemporary CDS. Then, in its second part, the masterclass will focus specifically on the DHA by highlighting its key ideas, distinctive features and standard DHA research designs and levels/categories of analysis. The former and the latter will be introduced during a hands-on workshop with students conducting group-work analyses subsequently discussed among the wider group. Finally, in its third part, the masterclass will highlight various research designs and projects developed while conducting DHA-based analyses tracing a variety of ‘discursive shifts’ in the public, political and policy communication. These will be explored by focussing on, on the one hand, research across various sites and genres of mediation and communication on dynamics of contemporary facets of far-right politics & illiberalism and its discursive strategies of normalising the politics of exclusion in relation to various forms of ‘crises’ (e.g. the 2015-16ff ‘Refugee Crisis in Europe’). On the other hand, the masterclass will also highlight the so-called Discourse-Conceptual Analysis (or DCA) drawing on DHA to explore the role of various mobilising social and political concepts in initiating and driving various ‘discursive shifts’ in contemporary political, policy and public including media discourse.

Background Reading:

Part I: CDA & CDS

• Krzyżanowski, M. (2010). The Discursive Construction of European Identities. Frankfurt am Main: Peter Lang. (Chapter 2: Dimensions of Analysis, Sections 1 & 2 – pp. 67-80).
• Krzyżanowski, M. & B. Forchtner (2016) Theories and concepts in critical discourse studies: Facing challenges, moving beyond foundations. Discourse & Society 27(3), 253-261. [https://journals.sagepub.com/doi/full/10.1177/0957926516630900]

Part II: DHA

• Krzyżanowski, M. (2010). The Discursive Construction of European Identities. Frankfurt am Main: Peter Lang. (Chapter 2: Dimensions of Analysis, Section 3 – pp. 81-89).
• Krzyżanowski, M. (2016). Recontextualisations of Neoliberalism and the Increasingly Conceptual Nature of Discourse: Challenges for Critical Discourse Studies. Discourse & Society 27(3). [https://journals.sagepub.com/doi/10.1177/0957926516630901]
• Krzyżanowski, M. (2018). Social media in/and the politics of the European Union: Politico-organizational communication, institutional cultures and self-inflicted elitism. Journal of Language & Politics 17(2): 281-304. [https://www.jbe-platform.com/content/journals/10.1075/jlp.18001.krz]
• Reisigl, M. 2018, The Discourse-Historical Approach. In J. Richardson and J. Flowerdew (eds) The Routledge Handbook of Critical Discourse Studies. London: Routledge, pp. 44-59. [https://www.routledgehandbooks.com/doi/10.4324/9781315739342.ch3]

Part III: (Analysing) Discursive Shifts

• Krzyżanowski, M. (2018), Discursive Shifts in Ethno-Nationalist Politics: On Politicization and Mediatization of the “Refugee Crisis” in Poland. Journal of Immigrant and Refugee Studies 16: 1-2. https://www.tandfonline.com/doi/full/10.1080/15562948.2017.1317897
• Krzyżanowski, M. (2020). Discursive shifts and the normalisation of racism: imaginaries of immigration, moral panics and the discourse of contemporary right-wing populism. Social Semiotics 30:4. [https://www.tandfonline.com/doi/full/10.1080/10350330.2020.1766199]
• Krzyżanowski, M. & N. Krzyżanowska (2022). Narrating the ‘new normal’ or pre-legitimising media control? COVID-19 and the discursive shifts in the far-right imaginary of ‘crisis’ as a normalisation strategy. Discourse & Society https://journals.sagepub.com/doi/10.1177/09579265221095420
• Krzyżanowski, M., R. Wodak, H. Bradby, M. Gardell, A. Kallis, N. Krzyżanowska, C. Mudde & J. Rydgren (2023). Discourses and practices of the ‘New Normal’: Towards an interdisciplinary research agenda on crisis and the normalization of anti- and post‐democratic action. Journal of Language & Politics 22(4): 415-437. [https://doi.org/10.1075/jlp.23024.krz]
• Ekström, H., M. Krzyżanowski & D. Johnson (2023). Saying ‘Criminality’, meaning ‘immigration’? Proxy discourses and public implicatures in the normalisation of the politics of exclusion, Critical Discourse Studies, https://doi.org/10.1080/17405904.2023.2282506
• Krzyżanowski, M. (2019). Brexit and the Imaginary of ‘Crisis’: A Discourse-Conceptual Analysis of European News Media. Critical Discourse Studies 16(4). [https://www.tandfonline.com/doi/full/10.1080/17405904.2019.1592001]

BIO: Professor Michał Krzyżanowski holds the Chair in Media and Communication Studies at Uppsala University, Sweden, where he is currently Deputy Head at the School/Department of Informatics and Media as well as Director of Research at the Uppsala University Centre for Multidisciplinary Studies on Racism (CEMFOR). He is one of the leading international scholars working on critical discourse studies of normalisation of the politics of exclusion in the context of illiberalism, far-right and neoliberalism. His focus is on communication, media and social change, specifically on European anti-immigration rhetoric, racism, social inequality and related challenges to liberal democracy. He is the Editor-in-Chief of the international Journal of Language and Politics and a co-editor of the Bloomsbury Advances in Critical Discourse Studies book series as well as sitting on a number of boards in various journals in critical discourse studies and wider qualitative social research. He is widely known for his teaching of qualitative methods and critical discourse studies to students across social and political sciences and humanities across Europe, Australia, China and USA. He is a recurrent convenor of widely attended classes in Critical Discourse Studies organised within Summer/Winter Schools in Methods and Techniques and Virtual Methods Schools of the European Consortium for Political Research (ECPR). More information: https://katalog.uu.se/profile/?id=N20-1042

Virginie van Ingelgom, Université Catholique de Louvain

How are you coding?
Aductive coding: From its epistemology to a set of tactics for qualitative coding of textual data

In current research, qualitative data includes a wide range of materials (e.g., newspapers articles, blogs, observations and fieldnotes, face-to-face interviews, focus groups). Coding qualitative data is a fundamental process that involves categorizing and organizing textual information to identify patterns, themes, and concepts. The process of coding helps researchers – as well as data analysts, consultants, journalists, students – to extract meaningful information from qualitative data and to draw conclusions. It is thus a key process to construct our knowledge and to tackle societal challenges.

In qualitative data analysis aiming at theory building, many researchers anchor their methodological approach – explicitly or not – in grounded theory as developed by Glaser and Strauss (1967). Grounded theory offers a useful set of steps for coding and analysing qualitative data inductively and has generated rich and insightful bodies of work. Yet, recent research argues that grounded theory’s central prescription – theoretical sampling toward saturation, strongly inductive analysis, and full immersion in the research field – bear little resemblance to the actual practices of researchers (Deterding & Waters 2018; Vila-Henninger, L., Dupuy, C., Van Ingelgom, V. et al., 2022).

Indeed, the grounded theory approach was developed based on projects with a relatively small number of interviews, generally conducted, and analysed by the researcher themselves. Increasingly however, qualitative studies involve large amount of qualitative data – referred to in the literature as Big Qual (Davidson et al., 2019). Moreover, today’s large-N studies often frequently involve teams of people who code based on a shared pool of material.

The development and uses of Qualitative Data Analysis (QDA) softwares – e.g., Atlas.ti, Delve, MaxQDA, NVivo – have helped a lot logistically. However, despite important technological developments, the coding procedures for theory building still mainly build on grounded theory principles and, thus, rely strongly on inductive analysis. Deduction is of course not absent of coding practices. However, deductive coding does not aim at theory building, as it consists of a top-down approach where qualitative analysis starts by developing a codebook with an initial set of codes based on existing analytical frameworks or theories.

From this discrepancy between research principles and new practices arises the issue of what avenues are to be taken to analyse and code qualitative data when research practices do not match anymore the central prescriptions of grounded theory. How should twenty-first-century qualitative researchers proceed in their qualitative coding?

This masterclass offers a crash introduction to how qualitative coding and abduction in general – and our particular method of abductive coding – provides a fruitful avenue for qualitative analysts. The application of abduction to qualitative research arose in the context of intense debate over the use of induction versus deduction. In this context, Timmermans and Tavory (Timmermans & Tavory 2012; Tavory & Timmermans 2014, 2019; Tavory 2016) advanced the strategy of abduction to move beyond the induction versus deduction debate. Abduction is then part of a movement that highlights what its proponents see as a false dichotomy between inductive and deductive analysis (Wagner-Pacifici, et al. 2015). The masterclass highlights how deploying abduction provides both an epistemology and a method of analysis that is adapted to many contemporary coding research practices of qualitative scholars. Our approach applies Timmermans and Tavory’s three stages of abduction— Revisiting the Phenomenon, Defamiliarization, and Alternative Casing—in three steps for qualitative analysis: Generating an Abductive Codebook, Abductive Data Reduction through Code Equations, and In-Depth Abductive Qualitative Analysis. This workshop aims to outline a set of tactics for abductive coding developed in the framework of our ERC Project Qualidem and to discuss with participants their own practices of coding qualitative data (Vila-Henninger, L., Dupuy, C., Van Ingelgom, V. et al., 2022; Dupuy & Van Ingelgom, 2024).

The masterclass will be divided into three parts. The first part will deal with a general introduction to abduction by focusing on its key assumptions, in relation to induction and deduction, while offering a quick overview of the key approaches in contemporary qualitative coding practices. Then, in its second part, the masterclass will focus specifically on abductive coding by highlighting its key ideas, distinctive features, and a set of tactics for abductive coding. Finally, in its third part, the masterclass will highlight various research designs and projects developed while conducting abductive coding. These will be explored by focusing on, on the one hand, research across different disciplines in social sciences and beyond. On the other hand, there will be also a moment of exchanges on researchers’ coding practices to open a dialogue on good practices and to detect who is doing abductive coding without knowing it. To prepare this third moment of the workshop, participants are invited to send a sample of their coding practices to Virginie Van Ingelgom: virginie.vaningelgom@uclouvain.be, with their potential questions on their coding practices, one week prior to the Masterclass.

Reading:

Abduction

• Tavory, I. (2016). The Pragmatist Wave of Theory Construction, The British Journal of Sociology, 67(1): 50-6. https://doi.org/10.1111/1468-4446.12187
• Tavory, I. & Timmermans (2022), Data Analysis in Qualitative Research. Theorizing with Abductive Analysis. Cicago: University of Chicago Press.
• Timmermans, S. & Tavory, I. (2012). Theory Construction in Qualitative Research From Grounded Theory to Abductive Analysis, Sociological Theory, 30(3):167-86. https://doi.org/10.1177/0735275112457914

New avenues in coding practices

• Deterding, N. M., & Waters, M. C. (2021). Flexible Coding of In-depth Interviews: A Twenty-first-century Approach. Sociological Methods & Research, 50(2), 708-739. https://doi.org/10.1177/0049124118799377
• Edwards, R., Davidson, E., Jamieson, L. et al. (2021), Theory and the breadth-and-depth method of analysing large amounts of qualitative data: a research note. Qual Quant, 55, 1275–1280. https://doi.org/10.1007/s11135-020-01054-x
• Vila-Henninger, L., Dupuy, C., Van Ingelgom, V., Caprioli, M., Teuber, F., Pennetreau, D., Bussi, M., & Le Gall, C. (2022). Abductive Coding: Theory Building and Qualitative (Re)Analysis, Sociological Methods & Research. https://doi.org/10.1177/00491241211067508

Abduction and abductive coding in practices

• Brandt, P. & Timmermans S. (2021). Abductive Logic of Inquiry for Quantitative Research in the Digital Age. Sociological Science 8: 191-210. http://dx.doi.org/10.15195/v8.a10
• Dupuy, C. & Van Ingelgom, V. (2024). Neither responsive, nor responsible? Citizens’ understandings of political actors’ responsiveness and responsibility in the socio-economic governance of the EU. Journal of European Public Policy, 31:4, 1126-1152. https://doi.org/10.1080/13501763.2023.2288236
• Theiss M. (2023). How Does the Content of Deservingness Criteria Differ for More and Less Deserving Target Groups? An Analysis of Polish Online Debates on Refugees and Families with Children. Journal of Social Policy, 52(4): 962-980. https://doi.org/10.1017/S0047279422000058

BIO: Virginie Van Ingelgom is a Senior Research Associate Professor F.R.S. – FNRS at the Institut de Sciences Politiques Louvain-Europe, UCLouvain. She is also a visiting professor at the European Political and Governance Studies Department at the College of Europe. She received her PhD from Sciences Po Paris and UCLouvain (2010). Her research interests focus on the issue of democracy and legitimacy at the subnational, national, European, and global levels, on indifference and ambivalence in citizens’ attitudes, on policy feedback and in qualitative and mixed methods. Her previous work has been awarded with the Jean Blondel Ph.D. Prize by the European Consortium for Political Research (2012) and an ERC Starting Grant for her project QUALIDEM – Eroding Democracies. A qualitative (re-)appraisal of how policies shape democratic linkages in Western democracies (2017-2023). She has been a visiting fellow at Oxford University (2009 and 2011), at Sciences Po Paris (2006-2010 and 2017), at Université de Montréal (2010), at the European University Institute (2022) and an invited Professor at Sciences Po Bordeaux (2018 and 2022), at Sciences Po Grenoble (2016, 2017 and 2018) and at the Graduate Institute Geneva (2020 – online). In 2024, she was awarded with an ERC Proof of Concept Grant for her project DACAT – Developing an Abductive Coding Analysis Toolkit.

Emilia Wiśnios & Bartosz Pieliński, University of Warsaw

(no information available at the moment)