Russian Contentious Event Dataset (RCED)

RCED Codebook This codebook describes the fields in the dataset file (identical CSV and XLSX formats are provided), as well as the procedures and definitions used for the additional classification. This classification is presented as an example of how the dataset can be used and as a methodological step undertaken for an analysis using RCED. In addition to the limitations outlined in the main article, this codebook also articulates the limitations related to the employed classification. Fields of the main RCED dataset date This field represents the date of an event reported in the row. The date is based on the content of the tweet whenever present. If the date was not present in the original tweet or a combination of tweets that referred to the same event, it is derived from the date of the original post. Some of the dates, where a limited number of tweets was found, can be inaccurate by approximately three days. city The name of the city, village, or other populated area that was described in the tweet. region The region where the event took place, derived from the name of the city using the Google Geolocation API in 2024 and confirmed using Google Gemini 1.5 Pro. Inaccuracies are possible where the name of a village or selo is not unique to a specific region; in such cases, preference was given to the larger locality. city_lat Latitude value of the given city for user and visualisation purposes. city_long Longitude value of the city for user and visualisation purposes, derived from Google Geolocation API and manually confirmed using Google Maps. region_lat Regional latitude for user and visualisation purposes, derived from Google Geolocation API and manually confirmed using Google Maps. region_long Regional longitude for user and visualisation purposes, derived from Google Geolocation API and manually confirmed using Google Maps. 1 protest_summary A summary of the event descriptions derived from the original tweets, particularly where multiple posts were observed about a similar event. It usually captures all information pertaining to the repertoire, the number of attendees, the topic, and the main political figures/organisations present. Due to a lack of contextual information, some descriptions are more general. The summaries were generated using OpenAI GPT-3.5 as the optimal choice for this summarisation task at the time of the dataset’s creation. Fields of ANTIGOV, PROGOV, and NONPOL datasets The additional datasets based on the data from the main RCED file were created using Large Language Models performing tasks such as Named Entity Recognition and zero-shot classification. The purpose of generating these additional classifications was to demonstrate the capacity and level of accuracy of the RCED dataset, as well as to illustrate what can be achieved using modern LLMs in the context of PEA. The definitions and event categories were created inductively based on the reported events and their content. Named Entity Recognition was performed using a Gemini model, while classification was performed using a fine-tuned GPT-3.5 model. The model was chosen based on a set of objective (Kappa and raw accuracy score) and subjective assessments of its output and accuracy in comparison with Google Gemini against a manually classified gold standard of 1,000 dataset entries. The classification used the definitions below: 1. ANTIGOV: Events targeted at the government and its representatives, and events that are organised or headlined by prominent opposition activists. Such events may oppose particular reforms, the war in Ukraine, the annexation of Crimea, infringements on civil liberties, or corruption, or address bad socioeconomic policies, including rallies against local officials, etc. CIVLIB Events related to issues of civil liberties, such as limiting freedom of speech or assembly, among others, including law enforcement violence against citizens. CORRUPTION Events related to the issue of corruption by officials and their organisations. ELECTIONS Events related to elections, their conduct, organisation, and results. GOVERNMENT Events related to general claims against particular leaders or parties, such as Putin or United Russia. LOCAL Claims for the resignation of local officials and other issues related to regional and city-level politics. 2 POLPRIS Events related to political prisoners and the detentions of political activists or unfair persecution/prosecution. REFORM Events related to reforms and legislation amendments, such as the pension reform, the reform of the academy of sciences, or constitutional amendments, among others. SOCECON Anti-regime claims made due to unsuccessful economic and social policies and grievances. WAR Events related to the invasion of Ukraine, annexation of Crimea, and other military operations overseas. 2. PROGOV: Events that are organised in support of the Russian government, including events supporting its political decisions, supporting the war in Ukraine, rallies organised by nationalists, and rallies against international pressures on Russia, etc. FOREIGN Rallies related to international politics and relations favouring Russian decisions and allies. GOVERNMENT Rallies demonstrating support for the Russian political regime, the President, the United Russia party, and pro-regime leaders and institutions. This encompasses rallies for stability, anti-Maidan and anti-Orange Revolution sentiments, election rallies supporting the regime, and rallies opposing the political opposition. LOCAL Rallies supporting local officials and organisations affiliated with the regime and the ruling party, such as mayors and regional representatives. NATIONALIST Rallies promoting Russian nationalism, emphasising Russian exceptionalism, and advocating for Russian unity. This includes rallies supporting Russian-speaking populations in other countries and languages (e.g., Ukraine), ’Russian brotherhood’, anti-Ukrainian sentiment, and rallies targeting specific nations (e.g., anti-American or anti-West rallies). WAR Rallies supporting Russian military operations, including the invasion of Ukraine, the war in Syria, and broader military actions. 3. NONPOL: Event reports that do not explicitly state whether the event was in support of or against the government, but that are related to specific issues such as the environment, commemoration events, mass celebration events, rallies for local benefits, infrastructure, or construction issues, etc. 3 ANIMAL Rallies advocating for animal rights, animal welfare, and the ethical treatment of animals. CELEBRATION Festive events marking national holidays (e.g., Victory Day, Labour Day), religious holidays, and other cultural celebrations. COMMEMORATION Solemn gatherings and events commemorating writers, politicians, historical events (e.g., withdrawal of troops from Afghanistan), anniversaries, and memorials. EDUCATION Rallies addressing issues related to education at all levels, including school funding, university access, curriculum content, and teacher rights. ENVIRONMENT Rallies and protests addressing environmental issues, including pollution, climate change, construction of controversial projects (e.g., incineration plants), and conservation efforts. ETHNIC Rallies organised by or in support of ethnic minority groups, promoting their cultural heritage, rights, and interests (e.g., Cossack rallies). FAMILY Rallies supporting traditional family values, children’s rights, and familyoriented policies. HEALTHCARE Events related to healthcare access, quality of medical services, and public health concerns. HOUSING Events focused on housing-related concerns, including mortgage borrower (defrauded shareholders) protests, construction and demolition issues, illegal development, land rights, living conditions, and utility access. INFRASTRUCTURE Rallies and protests concerning transportation and infrastructure, such as parking fees, tolls, public transportation quality, electricity supply, and water access. LABOUR Rallies organised by labour unions, and protests related to worker rights, wages, working conditions, and employment issues. MEDIA Rallies in support of media freedom, specific media organisations, or journalists facing pressure or censorship. RELIGION Events related to religious issues, such as the construction of religious buildings (e.g., churches), religious processions, and freedom of religion concerns. SPORT Events supporting sports teams, athletes, or specific sporting events. It is worth noting that some events may overlap, particularly in instances where protests encompass a variety of topics. In such cases, the first topic was chosen as the defining theme 4 for classification purposes. There might be occasional inaccuracies in defining some events due to the subjectivity of these classes, false outputs from GPT-3.5, and the lack of broader context within the tweets and descriptions. Tweets that contained insufficient information or clarity regarding their specific purpose were attributed to the broader "OTHER: rallies that do not fit into any of the above categories" category as outliers. COMMEMORATION and CELEBRATION events in the NONPOL category are categories that do not fit within the traditional definitions of contentious action. However, they are arguably important in instances where such events are organised by federal and regional governments, as they may incorporate pro-government narratives or be arranged in support of specific actions, such as the Crimea annexation or military celebrations during the war in Ukraine that started in 2022. Such events can be excluded depending on the purposes of the analysis. Other fields of the dataset that reflect the content, quality, and possible information obtainable through this dataset include: location The name of a location within the city, where specified and successfully identified by the language model (e.g., Pervomayskaya Square or Drama Theatre). If not present, this field is set to NONE. event_reason The reason why an event occurred or grievances raised by the protesters (e.g., construction of a silicone plant, or a memorial rally dedicated to victims of political repression). If no reason is identified, this field is set to NONE. mentioned_individual People mentioned in the context of the protest, where present or correctly identified by the LLM. NONE otherwise. mentioned_organisations Organisations mentioned in the context of the tweet or summary, where present or correctly identified by an LLM. NONE otherwise. participants Populations present at the event (e.g., "Airport staff" or "Deceived pensioners"). NONE otherwise, if unidentified or not present. These fields can be used for qualitative analyses, for triangulating events, or for confirming potential unaddressed duplicates. 5