In this talk, we present a hybrid methodological approach for studying large-scale digital discourse by combining computational analysis with close and interpretive reading. We use our recent study of the digital afterlives of lynching on Twitter as a case study to demonstrate how computational methods can help researchers identify patterns across large collections of social media data while retaining the contextual and interpretive depth of qualitative analysis.
We introduce the idea of scalable reading, which combines the breadth of computational or distant reading with the depth of close reading. Our dataset consists of 133,370 English-language tweets referring to lynching in the United States between 2007 and 2022. We demonstrate how different computational techniques can be used at different stages of the research process, including natural language processing, transformer-based named entity recognition, entity verification using Wikidata and Wikipedia, hashtag co-occurrence analysis, and BERT-based topic modeling.
Rather than treating computational results as the final interpretation, we use them to identify patterns, connections, and points of tension that can then be examined through close reading. For example, hashtag co-occurrence analysis allows us to trace how references to historical figures and events become connected with contemporary events and movements, while topic modeling reveals different and sometimes competing meanings associated with the term “lynching.”
We will introduce three methodological concepts that emerged from this combination of computational and interpretive analysis: temporal traction, referring to how digital references connect past and present; semantic abrasion, referring to the collision of competing meanings and interpretations; and visibility routing, referring to the uneven ways in which platform infrastructures shape what becomes visible, by whom, and where.
The talk will also discuss the methodological challenges of working with large-scale social media data. Computational approaches can reveal patterns that would be difficult to identify through close reading alone, but scaling up does not eliminate the need for contextual knowledge, interpretation, and ethical reflection. In particular, we discuss the risks of decontextualization and the ethical challenges involved in computationally studying histories of racialized violence.
Finally, we reflect on what this approach can offer researchers working with other forms of large-scale textual or digital data. The central methodological question is not whether computational or qualitative methods are preferable, but how the two can be combined so that computational patterns become starting points for interpretation rather than substitutes for it. This approach offers a way of working across disciplinary and methodological boundaries and may be useful for researchers in the humanities and social sciences who want to incorporate computational methods into the study of digital texts, discourse, memory, and cultural phenomena.
Feeza Vasudeva is an Academy Researcher at the University of Helsinki (Research Council of Finland, 2025–2029), leading the DIVINE project (Digital Interfaces and Virtual Innovations in New Expressions of Faith). Her research investigates how digital technologies reshape political belief, religious publics, and democratic life. Her work spans multiple interconnected areas: as part of HSSH's ‘Datafication of Society’ initiative, she examined how collective violence, specifically lynching in India and America, circulates, persists, and fragments across digital platforms; simultaneously, she investigated political deification and religious populism in global contexts. Within DIVINE, this now extends to AI-generated religious content and aesthetics, including her work on Godbots, a study of how conversational AI reconfigures who or what can speak for the sacred.
Narges Azizifard is a PhD in Computer Science and a researcher with experience in machine learning, natural language processing, and digital humanities. She worked as a Postdoctoral Researcher at the Department of Digital Humanities, University of Helsinki (2022–2025), where she used NLP and machine learning methods to analyze large-scale social media data, including Reddit and Twitter. Her research interests include computational social science, spatial analysis, digital humanities, NLP, and data-driven discourse analysis.
Eetu Mäkelä is a professor of Digital Humanities (Human Sciences–Computing Interaction) at the University of Helsinki. At the Helsinki Centre for Digital Humanities, he leads a research group that seeks the technological, processual and theoretical underpinnings of successful computational research in the humanities and social sciences. Additionally, he serves as a technological director at the DARIAH-FI infrastructure for computational humanities and is one of three research programme directors in the datafication research initiative of the Helsinki Institute for Social Sciences and Humanities. He also leads the Helsinki Liberal Arts and Sciences Bachelor’s Programme at the University of Helsinki.
In this talk, we present a research method that enables the automatic analysis of facial expressions from video recordings. In 2023 HSSH organized a workshop on the method that inspired its use in the context of language education and language assessment. We give two examples on how the method has been used in the Aasis project (Research Council of Finland 2023 – 2027) developing ways to automatically measure and assess verbal and nonverbal features of spoken interaction in L2 Finnish.
We briefly present a manual (Juselius et al. 2026) introducing the use of algorithm-based facial expression analysis in research that was published for researchers at the University of Helsinki City Centre Campus. The development of the manual was supported by HSSH’s Catalyst Grant funding. We will also explain what we have learnt about the limitations of the method, and what new possibilities the software released this summer, Py-Feat Live, offers (
The method offers new opportunities for analysing interaction and interview data, for example. It can also be used in research on online work, such as examining users’ facial expressions, depending on the type of video material the researcher has collected and the research questions the data are intended to address. With the help of the manual, University of Helsinki researchers can independently use the Py-Feat software to analyse videos recorded, for example, in Interlab (bookings via UH’s Outlook calendar).
Py-Feat enables algorithm-based analysis of facial expressions and can therefore complement researchers’ methodological toolkit, especially when the research focuses on interaction, reactions, or nonverbal communication using video data. Please let us know if you would like to take part in the workshop we are planning for late autumn.
Anna von Zansen, is a University Researcher working for two Academy Projects, Aasis (Research Council of Finland 2023 – 2027) and DigiTala in action (2025 – 2026) at the Faculty of Educational Sciences, University of Helsinki. Her research interests include educational technology, language assessment, multimodality, L2 listening and speaking, spoken interaction.
Joona Juselius, is a Research Assistant at the Faculty of Educational Sciences, University of Helsinki. He worked with funding from HSSH (Catalyst Grant 2026) and produced a manual on Py-Feat to support researchers at the city centre campus. He works in Anna von Zansen’s team and uses the method in his master’s thesis, majoring in English.
Mariel Wuolio, is a Project Planner at the Helsinki Institute for Social Sciences and Humanities. She works as the coordinator of the interaction research laboratory Interlab, administered by HSSH, being responsible for the laboratory’s services and user guidance.
References:
Juselius, J., Henttonen, P., Wuolio, M., & von Zansen, A. (2026). An addition to researchers’ toolkit: Algorithm-based facial expression analysis in L2 education (v1.0). Zenodo.