Since the beginning of Russia's full-scale invasion of Ukraine in 2022, public protest has become increasingly fragmented, short-lived, and vulnerable to erasure due to censorship, repression, and platform moderation. Much contemporary civic mobilisation risks disappearing from the historical record. This paper presents a public digital humanities project that addresses this challenge by developing an open, reproducible infrastructure for documenting and preserving protest activity in Russia through social media data.

The project is based on large-scale data collection from VKontakte and Telegram, the dominant social media platforms in Russia and key sites where information about protests circulates under authoritarian conditions. Using automated data extraction pipelines, the project has collected and processed posts referencing protest activity from 2022 onwards, resulting in a dataset documenting nearly 40,000 protest events across all Russian regions. The infrastructure (github.com/dariatalanova/russian-protests) was originally developed to support an investigative journalism project published by Novaya Gazeta Europe, but has since been redesigned as a reusable, open-source research framework.

Methodologically, the project combines computational text analysis with large language models (LLMs) to address challenges of scale, duplication, and semantic ambiguity in social media data. The pipeline includes automated scraping, text cleaning, vectorisation using embeddings, clustering and deduplication of semantically similar posts, and LLM-assisted entity extraction to identify locations, protest types, and organisational actors. Rather than treating LLMs as black-box analytical tools, the project frames them as infrastructural components that enable the transformation of unstructured, ephemeral social media content into a structured, archivable corpus.

The paper argues that such infrastructures function as critical knowledge systems in contexts where civic expression is actively suppressed. Protest events documented on social media are often rapidly deleted, obscured, or distorted; by systematically capturing and preserving these traces, the project creates a digital archive of civic mobilisation that would otherwise be lost. The infrastructure operates not only as a research tool but also as a mechanism for preserving collective memory, enabling future historical, sociological, and cultural analysis. By making both the methodology and the codebase openly available, the project supports multiple research futures: quantitative analysis of protest dynamics, qualitative case studies, comparative regional studies, and longitudinal investigations of civic activism under repression.

The project engages with ethical questions central to public digital humanities, including responsible handling of sensitive data, minimisation of harm to individuals, and the limits of transparency in authoritarian contexts. Decisions regarding data aggregation, anonymisation, and selective publication are embedded in the infrastructural design rather than treated as afterthoughts.

This paper contributes to ongoing discussions within digital humanities on research infrastructures as forms of public engagement, the preservation of fragile digital heritage, and the ethical use of computational methods in politically sensitive environments. It demonstrates how methods developed at the intersection of journalism, computational social science, and digital humanities can build sustainable, open infrastructures that document civic life under repression, supporting both scholarly inquiry and the public good.