Large language model analysis reveals key reasons behind massive farmer protests in Europe

Large language model analysis reveals key reasons behind massive farmer protests in Europe | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Large language model analysis reveals key reasons behind massive farmer protests in Europe Christian Stetter, Eva-Marie Meemken, Andrea Fürholz, Robert Finger This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6652927/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract European farmers face mounting pressures. They must produce food while navigating climate change and preserving ecosystems. In 2023/24, these tensions erupted into massive protests across the continent. Despite their importance, farmer protests in Europe have received little attention and its drivers remain largely unclear. We address this gap by using large language models to analyze comprehensive data on 4,642 farmer protests throughout Europe from November 2023 to March 2024, and assess the main reasons for these protests. We find a complex interplay of economic, political and social pressures. Key reasons include opposition to foreign imports, resistance to EU policies, and frustration over subsidy cuts and delays. The analysis highlights significant geographical and temporal variations, with major protest hotspots in France, Germany, Spain, and Poland. Our findings highlight the need for solutions to balance environmental sustainability with profitability and ensuring a just transition for all stakeholders in the agrifood system. Scientific community and society/Agriculture Scientific community and society/Social sciences/Interdisciplinary studies Figures Figure 1 Figure 2 Figure 3 Figure 4 Main text The agricultural sector, in Europe and worldwide, is facing unprecedented challenges: it must produce sufficient and diverse food for all, without exacerbating climate change and other environmental challenges, while ensuring farmers can earn a sustainable livelihood (Mustafa et al., 2024; Schebesta & Candel, 2020; Tilman et al., 2011; Yang et al., 2024). These issues have been further intensified by recent external shocks such as the COVID-19 pandemic and the war against Ukraine, which have disrupted global supply chains, amplified uncertainty, and increased the demand for greater food self-sufficiency (Abay et al., 2023; Behnassi & El Haiba, 2022; Clapp & Moseley, 2020). In Europe, farmers receive generous subsidies (e.g., 30% of the EU budget) designed to balance agricultural productivity, environmental protection, and farm profitability (European Parliament, 2023; Foley et al., 2011; Pe’er et al., 2019). Yet, farmers feel under immense pressure, believing that the support they receive is inadequate to address the challenges they face. This strain has culminated in massive farmer protests across Europe in 2023/24. Farmers took the streets to voice their frustration with the current situation and demanded action (Finger et al., 2024). Many farmers felt their livelihoods are at risk, struggling to balance environmentally sustainable practices with economic survival. Although farmer protests have a long-standing history in Europe and elsewhere (Bush & Simi, 2001; De Weerd & Klandermans, 1999; Naylor, 1994), the protests in 2023/24 were unique in their magnitude and geographical spread—and have drawn widespread media attention. This led to prompt reactions from policy makers, significantly contributing to the reversal of ambitious environmental goals in European agriculture (Matthews, 2024b). To fully understand the dynamics of these protests and their potential impact on European agricultural policy, it is key to understand the underlying reasons of the protests and their development over time and space. The extant literature on the 2023/24 farmer protests is extremely limited. So far, research on these farmer protests have been based on anecdotal evidence and unsystematic evaluations of media reports (Finger et al., 2024; Matthews, 2024b, 2024a). These studies have highlighted pressure on farm incomes, tighter environmental regulations, and trade competition (Matthews, 2024a) as possible overarching drivers of farmer protests, which can differ across countries (Finger et al., 2024). However, systematic analyses of these protests are absent, and so is evidence on protest reasons across time and space. This study is the first to provide a comprehensive and systematic analysis of reasons and struggles driving the 2023/24 European farmer protests. We use data from the Armed Conflict Location & Event Data Project (ACLED), an extensive database capturing more than 4,600 farmer protest events across Europe from November 2023 to March 2024. The database contains location, date and qualitative information on individual protests and underlying themes and motivations (Raleigh et al., 2010, 2023). For our analysis, we leverage the capabilities of large language models to identify important protest reasons and classify farmer protest events accordingly (Ash & Hansen, 2023; Gilardi et al., 2023; Srivastava et al., 2022). Modern large language models (LLMs), such as GPT (OpenAI et al., 2023) have shown exceptional performance in various natural language processing tasks, including translation, question-answering and text classification, often surpassing human performance (Ibrahim et al., 2023; Mittelstädt et al., 2024; Törnberg, 2023; Webb et al., 2023).We use LLMs to process the large volume of text data (i.e. qualitative descriptions of the 4,303 farmer protest events) to achieve two goals: i) to identify key reasons for farmer protests and ii) to classify protest events into categories aligned with these identified reasons. We apply this analysis over time, capturing the period from 1 November 2023 to 31 March 2024—the main period of the protests—and across space (i.e., covering all European countries). This allows us to uncover and quantify trends and patterns in the evolution of farmer protest movements, as well as regional or temporal variations. A diverse set of reasons were behind the farmer protests Using the large corpus of available protest descriptions reported in the ACLED dataset, and employing an LLM-human collaborative approach (see Online methods), we detected 18 protest reasons that are described in Table 1 . Reasons include a variety of interlinked and sometimes overlapping issues, such as (i) opposition to farm, trade, and environmental policies as well as (ii) dissatisfaction with market developments, prices, working conditions, and livelihood opportunities. Table 1 LLM-human detected reasons for the European farmer protests from November 2023 - March 2024 Protest reasons Descriptions 1 Rising production costs Protests driven by the increasing costs of production, including fuel, fertilizers, insurance, and energy, and other agricultural inputs, impacting farm profitability and contributing to financial pressures on farmers. 2 Opposition to EU free-trade agreements Protests and demonstrations against the ratification of the EU's free-trade agreements, demanding that partner countries comply with EU agricultural standards to ensure fair competition and protect local farmers' interests and maintain market stability. 3 Environmental regulations and agricultural standards Protests against various agricultural policies set by the European Union, including the Green Deal, and Common Agricultural Policy. Protesters oppose stricter environmental norms and regulations, particularly those related to pesticide and nitrogen use, which are perceived as overly restrictive and detrimental to agricultural practices. 4 Bureaucratic constraints Protests against excessive bureaucratic processes and administrative burdens related to agricultural operations, state support, and subsidies that farmers believe hinder operational efficiency, productivity, sustainability in the agricultural sector. Protesters lament complex regulatory frameworks, excessive paperwork, and compliance requirements. 5 Opposition to foreign agricultural imports Demonstrations and actions targeting the import of agricultural products from foreign countries, particularly Ukraine, which are perceived to undercut local market prices and harm domestic farmers. The movements emphasize the need for fair production prices and express concerns over the quality of imported goods, which are seen as detrimental to consumer trust and local agricultural markets. 6 Subsidy cuts Protests against government plans to cut or delay subsidies and tax exemptions critical to the agricultural sector. These protests often focus on issues such as reductions in agricultural diesel subsidies, vehicle tax exemptions, and overall financial support necessary for farm operations. The protests reflect a broader demand for increased state support and highlight the economic challenges faced by the agricultural community due to inadequate subsidies and budget cuts. 7 Climate and natural disaster relief Protests and movements advocating for relief measures and compensation for farmers to offset the consequences of droughts, floods, and other climate-related natural disasters impacting agricultural production. These movements emphasize the urgent need for financial support, sustainable practices, and specific policies to mitigate the adverse impacts of climate change and environmental challenges on agricultural production. 8 Labor and social conditions Movements and protests advocating for improved working and social conditions for farmers and farm workers, including disputes over working hours, fair pay, safety, financial security, health, and support measures during crises. They also call for greater recognition and appreciation of their work. 9 Solidarity movements Demonstrations showcasing support among farmers and other sectors, jointly advocating for farmers' recognition, and addressing shared economic challenges, regulatory burdens, and broader social issues such as anti-racism, immigrant rights, and social equality across Europe. 10 Opposition to renewable energy projects Demonstrations opposing large-scale renewable energy projects, such as wind and solar parks, due to concerns about their negative impact on agricultural land, practices, and local environments. 11 Economic struggles and agricultural livelihoods Protests addressing the economic challenges faced by farmers, including low production prices, high costs, debt relief and demands for financial support, better working conditions and survival of the agricultural sector. 12 Consumer awareness initiatives Efforts to raise awareness among consumers to educate the public on agricultural issues and the effects of policies on food production and pricing, often involving direct actions at retail locations. 13 Livestock and animal welfare protests Protests focused on the challenges faced by livestock breeders, including concerns over animal welfare regulations, insufficient financial support, market conditions, and the impact of EU policies on livestock and beekeeping, that are seen as detrimental to farmers' livelihoods. 14 Opposition to non-traditional products Protests against laboratory-made meat, insect-based flour, and other novel agricultural products perceived to threaten traditional farming practices and cultural authenticity. 15 Infrastructure and transport policies This category encompasses various demonstrations aimed at addressing grievances related to infrastructure and trade policies. Key issues include unfavorable trade and transport policies affecting transport companies, particularly in cross-border scenarios, as well as local environmental concerns. The protests frequently involve stakeholders such as truck drivers, transport unions, and local communities advocating for fair labor regulations and sustainable infrastructure development. 16 National and local state support Protests and demonstrations focused on the need for increased state financial support and subsidies for the agricultural sector, addressing issues like low compensation, economic difficulties, and specific local and national policies affecting farmers. 17 Fair compensation and market practices Protests centered around low prices paid for agricultural products, which are seen as insufficient to cover the costs of production and ensure a viable income for farmers, protection for local farmers against low-cost foreign products, fairer compensation from large retailers, and addressing unfair trade practices that impact local farmers. 18 Miscellaneous agriculture-related protests Protests and demonstrations related to various miscellaneous aspects of the agricultural sector. It includes issues such as technological regulations, employment and training, public health and safety, immigration and trade, environmental sustainability, political demands, and unique local grievances. Note: These reasons were identified and categorized by leveraging GPT-4’s advanced natural language processing capabilities. By refining prompt engineering techniques, we generated a coherent classification of protest reasons, ensuring accuracy and contextual relevance. To address LLM limitations, we incorporated human refinement. These reasons and descriptions were provided to a GPT-4 to apply zero-shot multilabel classification to label each protest event during the European 2023/24 farmer protests according to the detected categories of reasons. By using state-of-the-art prompting strategies and validating the model against trained human annotators, we could ensure a coherent protest event classification. Details are provided in the Online methods and Supplementary Information (SI). Major driving forces and their spatial distribution across Europe Geographically, farmer protests surged across Europe, with particularly high levels of protest activity, concentrated in Germany (1,072 protests or 23% of all protests), France (932, 20%), Poland (877, 19%), Spain (576, 12%), and Italy (266, 6%). These countries contributed to 81% of all protests. Scandinavian countries, the UK, and the Western Balkans saw much lower protest activity (accounting for less than 2% of all protests), underscoring regional disparities in the numbers of agricultural protests across the continent (see central panel in Fig. 1 b, and Supplementary information SI 6). The main results, illustrated in Fig. 1 , reveal the prevalence of the reasons for the protests (Panel A). Although protest reasons are strongly interconnected (for example, subsidy cuts often co-occurred with concerns over rising production costs and market conditions, Fig. 1 c and SI 6), there are different patterns across countries, underlined by looking at the following three most prevalent reasons. First, opposition to foreign agricultural imports was the most common protest reason, having been associated with nearly 50% of protest events. This sentiment was especially strong in France, Poland, and Spain (see Fig. 1 B and Fig. 2 ), where imports—particularly from Ukraine—were perceived as harmful to domestic farmers due to their competitive pricing. Many protests highlighted concerns about the perceived lower quality of imported goods and the increased market competition they brought (Table 1 ). The second most prevalent protest reason (~ 30%) was opposition to various agricultural policies, including the Green Deal and Common Agricultural Policy of the European Union (Fig. 1 A), with particular emphasis in France, Italy, and Belgium (Fig. 2 ). Protesters expressed discontent with stricter environmental regulations, especially those limiting pesticide and nitrogen use, which were seen as overly restrictive and detrimental to agricultural productivity. Lastly, the third most common protest reason (~ 27%) involved opposition to proposed government cuts or delays in subsidies and tax exemptions, notably in Germany (Fig. 1 C). These protests often centered around reductions in subsidies for agricultural diesel, vehicle tax exemptions, and general financial support to farm operations. The protests reflect a broader demand for increased state support and highlight the economic challenges faced by farmers despite policy support. Other significant factors included rising production costs, bureaucratic constraints, and dissatisfaction with market conditions. The prevalence of reasons varied not only across countries, but also over time (Fig. 1 d), which is illustrated in more detail in the accompanying online application showing temporal and spatial heterogeneity: https://farmer-protest-app.onrender.com/ ). Overall, we find that two initial waves were provoked by subsidy cuts, then reasons diversified into broader concerns about trade, policy, and economic struggles, and protests spread throughout much of Europe. Discussion Agriculture is facing unprecedented challenges that demand a transformative shift toward greater sustainability. In this context, large-scale farmer protests across Europe in 2023/24 have emerged as a powerful expression of widespread dissatisfaction, revealing the multifaceted struggles currently felt by many European farmers. Our findings demonstrate that the reasons behind the European farmer protests were simultaneously heterogeneous, context-dependent, yet interconnected. The spatial distribution of the protests reveals important insights into the geography of farmers’ discontent. While France and the Netherlands have historically been focal points for farmer protests (De Weerd & Klandermans, 1999; Naylor, 1994), the farmer protests in 2023/24 marked a significant shift, characterized by high levels of protest events across the continent, and a contagious spread of these events in other parts of Europe, where such protests used to be less common. This shift suggests that farmer grievances have broadened geographically, indicating a growing transnational awareness of shared and mutual challenges. The geographical pattern of the protests reflects both the local specificity of farmers’ struggles and the growing interconnectedness of challenges European farmers are concerned with. For example, on the one hand, the European farmer protests of 2023/24 started locally in response to suggested cuts in agricultural diesel subsidies in Germany (Escritt, 2024), while in Poland, farmers’ discontent stemmed from rising production costs and fears over foreign agricultural imports, particularly from Ukraine (Finger et al., 2024). These differences highlight the importance of understanding local contexts, as farming systems across Europe operate under varying economic, social, and environmental conditions. Relatedly, not all regions exhibited the same level of farmer protests. Scandinavian countries, the United Kingdom, and parts of the Western Balkans reported relatively low protest activity. This unevenness raises questions about (i) whether some regions are better equipped to navigate agricultural challenges, which has in fact been found by Slijper et al. (2022) or (ii) if latent struggles remain unexpressed, e.g., due to weaker organizational capacity. On the other hand, the initial protests in Germany quickly gained traction in neighboring countries, as farmers recognized parallels in their own struggles. These developments illustrate how initial, localized struggles can escalate into transnational movements. This diffusion underscores how some of the contemporary struggles by farmers transcend national borders, fostering a shared sense of urgency among European farmers. Generally, the protest dynamics reveal a complex interplay of short-term shocks and long-term stresses. As Meuwissen et al. (2019) highlight, farming systems are shaped by these two categories of challenges (short-term shocks and long-term stresses). This dual framework demonstrates how short-term shocks can destabilize farming systems already weakened by long-term stresses, often leading to non-linear effects, heightened vulnerability, and an increased likelihood of tipping points. In the context of the protests, an acute policy shock—specifically the suggested Diesel subsidy cuts—triggered the initial demonstrations in Germany (Escritt, 2024). These quickly expanded to encompass a wide range of struggles reflecting (perceived) long-term stresses, such as administrative burden, red tape, rising production costs and competitive price levels. This evolution stresses why the protests became so widespread, exposing the deep frustrations that had been simmering beneath the surface long before the recent crises. Farmers view this eruption as an expression of the mounting pressures they have faced over time (Forst, 2024). Along these lines, Buitenhuis et al. (2020) show that the current common agricultural policy does not sufficiently support deeper levels of resilience within European agriculture. Farmers in high-income countries have historically been well-organized and powerful. The 2023/24 protests represent a new form of farmers leveraging their power in public debates, utilizing media attention, and influencing policy decisions, such as those on environmental protection. Relatedly, protests have already erupted again in the winter of 2024/25, including France, Spain and Belgium, among other countries (Macedonio Vega, 2024), indicating this new quality of influence is here to stay. More broadly, these dynamics, therefore, suggest that structural resistance to change remains a significant barrier for transformation in European agrifood systems. Many of the core reasons behind the protests are directly tied to farmers’ dissatisfaction with agricultural policies, and point to important goal conflicts between farm profitability, productivity, food self-sufficiency, and environmental protection – conflicts that these policies are designed to resolve but are not effectively solving in reality. Despite these existing tensions, farmer protests should not justify lowering environmental sustainability ambitions. While farmers perceive many regulations as burdensome and out of touch with their realities, policymakers face the challenge of balancing environmental imperatives with social and economic viability (Pe’er et al., 2019). Environmental regulations are essential to combat climate change and biodiversity loss (Pe’er et al., 2014, 2019), but the process of implementing such policies may have left many farmers feeling overwhelmed (Mennig, 2024). Ultimately, the protests highlight the need for a balanced approach to agricultural transformation. Previous research emphasizes the need to factor in the lived experiences of farmers when designing policies. Farmers’ self-concepts are often characterized by productivist and entrepreneurial identities and tied to their roles as stewards of the land (Burton & Wilson, 2006; Janker et al., 2021). Policy changes can threaten this self-concept or diminish their sense of purpose, not only leading to resistance but also threatening the social sustainability of European agriculture (Saleh & Ehlers, 2023). Future policy efforts must bridge the tensions between farmers’ subjective perceptions and objective realities by fostering stronger collaboration, communication, and coordination among policymakers and farmers (Mennig, 2024) — while also recognizing that transformations inevitably produce winners and losers, and thus offering new pathways or compensation for those whose livelihoods are threatened (Gaupp et al., 2021). Apart from that, this study demonstrates the value of large language models in the context of farmer protests and beyond. LLMs effectively classified farmer protest descriptions, offering new opportunities for understanding complex social phenomena. At the same time, this method has limitations. The data—short protest descriptions filtered by ACLED—capture only primary protest reasons, leaving secondary reasons, not reflected in the data, unexamined. Conflict and protest data, like that used by ACLED, often comes with inherent challenges related to reporting bias, which may stem from systematic inclinations, prejudices, and the directionality of information, influencing how these events are portrayed, how they are selected, and which aspects are emphasized (Miller et al., 2022; Weidmann, 2016). Although ACLED has strong quality control mechanisms (ACLED, 2023) that help reduce bias and maintain data reliability (see Online methods), it is impossible to completely eliminate bias from the process. Additionally, the LLM classifier cannot be tested against a ground truth, as the true nature of many events is inherently open to interpretation. This limitation, however, is not unique to machine-based classification; human annotators face the same challenge when classifying events, as their assessments are similarly influenced by context, perspective, and available information (Aroyo & Welty, 2015). To address this concern, we present evidence of agreement among human and machine annotators in the Supplementary Information. Finally, future research could complement our analysis with in-depth interviews, surveys, and qualitative methods to better understand the underlying dynamics, identify the types of farmers involved in protests and capture secondary reasons, farmer demographics, and quantify protest impacts on outcomes such as policy responses, voting behavior or market prices. Online Methods Overview This section outlines the methodology used to analyze farmer protests in Europe from November 1, 2023 to March 31, 2024, using data from the Armed Conflict Location & Event Data Project (ACLED) (Fig. 3 ). To categorize the reasons behind these protests, a hybrid approach was employed, leveraging the capabilities of large language models (LLMs) with human refinement. This process involved systematically analyzing textual descriptions from ACLED, generating protest reason categories through GPT-4, and validating these categories via human evaluation. Subsequently, an LLM-based zero-shot multi-label classification model was developed to assign (multiple) protest reasons to individual events, enabling a comprehensive understanding of the underlying reasons and patterns of these demonstrations. Data description The data used in this study comes from the Armed Conflict Location & Event Data Project (ACLED), a comprehensive database that specializes in collecting and analyzing information on political violence, protests, and conflicts in various regions of the world (Raleigh et al., 2010). ACLED records events in real time, providing detailed data on incidents ranging from armed conflict and political violence to demonstrations and other forms of political unrest. This database is central to understand the local and regional dynamics of crises (Raleigh et al., 2010). The ACLED database contains over 1.5 million recorded events categorized into six primary types: combat, violence against civilians, remote violence (e.g., bombings), riots, protests, and strategic developments (Raleigh et al., 2023). Events often involve multiple actors, including governments, rebel groups, militias, or other organized groups with political goals or effects. It should be mentioned that events with and without violence are included (Raleigh et al., 2023). Relatedly, ACLED defines a protest event as a (peaceful) in-person demonstration by three or more persons (here farmers), at a specific date and location opposing a political, governmental, institutional, or private entity, where violence is absent on their part but may be directed against them (ACLED, 2024). ACLED's data collection follows a methodology that draws from a wide range of sources, including international, national, and local news agencies, nongovernmental organizations (NGOs), government reports, social media, and direct reports from eyewitnesses and regional partners. This multi-source approach ensures comprehensive coverage, especially in underreported or remote areas, and provides a more complete picture of global unrest. Details on the methodology are described in detail in (Raleigh et al., 2023), highlighting the importance of using diverse sources to capture localized conflict. In addition, ACLED's partnerships with more than 50 local conflict observatories and its reliance on more than 13,600 sources in more than 100 languages ensure a broad and detailed representation of events (Raleigh et al., 2023). Validation processes include inter-coder, intra-coder, inter-code, and inter-source reliability checks, as well as algorithmic checks and manual reviews by analysts to ensure data accuracy and reliability. ACLED emphasizes the credibility of sources and makes its methodology and sources publicly available to promote scientific accountability and allow for external verification (Raleigh et al., 2023). The database's applications have been demonstrated in numerous studies across various domains. For instance, Anderson et al. (2021) and Deininger et al. (2023) used data to analyze conflict and food insecurity, while Maystadt & Ecker (2014) and O’Loughlin et al. (2012) explored climate variability and conflict risk. Plümper et al. (2021) and Shuman et al. (2022) examined protest dynamics related to COVID-19 and the BlackLivesMatter movement using ACLED. For this analysis, the ACLED database was carefully filtered according to specific criteria to ensure the data's relevance to the research objectives. The selection process targeted events within the defined geographic region and timeframe, focusing on conflict types and actors pertinent to our study. Specifically, the data were restricted to events occurring between November 1, 2023, and March 31, 2024. Additionally, the filtering criteria required that protests take place in Europe, involving either farmers or organizations associated with them. The data extraction process is illustrated in Fig. 4 . Class construction and descriptions To distill the reasons behind these protests, it was necessary to systematically process the large corpus of text information provided. To that end, we leveraged the power of large language models (LLMs). LLMs have shown impressive performance in a wide range of tasks, including problem-solving and reasoning, natural language understanding and generation, as well as text summarization and classification, and even surpasses human performance in many of these tasks (Ibrahim et al., 2023; Mittelstädt et al., 2024; Törnberg, 2023; Webb et al., 2023). In terms of text analysis, a big benefit of LLMs compared to humans is that they can systematically analyze large corpuses of text in a short time at low cost (Gilardi et al., 2023). To construct classes of farmer protest reasons, we prompted a large language model (“Proposer”), namely GPT-4 (OpenAI et al., 2023), with the goal of obtaining a coherent list of classes that explain the reasons for the protests and their corresponding class descriptions. While there are several open and closed language models, OpenAI’s GPT has performed consistently well in recent years and GPT-4 is a benchmark in many natural language processing tasks (Chiang et al., 2024). GPT (Generative Pre-trained Transformer) is a neural network model built on the Transformer architecture, trained on vast datasets of text and code to understand and generate contextually relevant responses. It uses self-attention mechanisms to process sequential data efficiently (OpenAI et al., 2023). GPT, at its core, is a statistical model to generate responses. This means it uses probabilities to decide which token (a word or part of a word) to generate next. It predicts that next token by calculating the probabilities of all possible tokens based on the preceding context in the sequence (Zhao et al., 2024). As previous research on LLMs has demonstrated, the quality of their output is highly dependent on prompt engineering—specifically, how the user frames their task for the model (Sahoo et al., 2024). To tackle this, we explored multiple prompt variations through an iterative, interactive process, guided by established best practices in prompt engineering (Sahoo et al., 2024). We accessed GPT via OpenAI’s API, refining our approach by testing different formulations and selecting the one that produced the best results. The final prompt used to generate categories of protest reasons, and their corresponding descriptions is presented in Table 2 . Table 2 Prompting strategy for suggestion of reason categories. System prompt You are Bob, an intelligent expert at categorizing textual information on farmer protests in Europe between 2023 and 2024. You are tasked with discerning key issues, a.k.a. categories, from textual descriptions and reasons of farmer protests. Make the categories diverse and distinguishable across, and coherent within. Output: You provide categories and a concise description for each category in table format only. User 1 prompt I will give you some text and you will analyze it. Got it? Assistant prompt Yes, I understand. I am Bob, and I will analyze the given task as best as I can. User 2 prompt Great! Let's begin then :). Analyze the following text: \n\n {text} \n Answer: Let's reach a comprehensive categorization. To address the inherent limitations of LLMs (Zhao et al., 2024), we incorporated human refinement to finalize the list of protest reasons and their corresponding descriptions. Initially, we divided the text corpus into three random subsets to ensure the content was manageable within GPT-4's token limits. For each subset, we ran five model iterations with the temperature hyperparameter set to one.[1] In total, 15 responses were generated, each consisting of categories and corresponding descriptions (see Supplementary information SI.1). Two authors independently sorted these responses and identified patterns and coherences across them. For example, all 15 models identified a variation of the category "Rising Production Costs." The sorted lists from both evaluators were subsequently compared (see Supplementary information SI.2). Based on this comparison, all authors reached a consensus on the final set of 18 labels and descriptions of the protest reasons. In a refinement step, the clarity and comprehensiveness of the proposed categories were evaluated by trained research assistants, who were tasked with labeling event descriptions according to the suggested protest reasons and descriptions. This process also served as a pre-test and training step for the subsequent classification task (see below). The feedback provided by the research assistants led to minor adjustments in the initially agreed-upon category descriptions to enhance clarity and avoid ambiguity. Multilabel classification model After establishing a list of potential protest reason categories in the initial step (“Proposer”), we employed multi-label classification to assign the relevant protest reasons to each protest event (“Multi-label Classifier”). Multi-label classification is a type of classification task where each instance can be associated with multiple classes, rather than just one. Unlike single-label classification such as binary and multiclass classification, where each instance is assigned only one class from a set of possible labels, multilabel classification allows for the prediction of multiple, simultaneous labels for a given instance (Abraham et al., 2009). This is the case in our application, where multiple reasons can simultaneously be associated with a single farmer protest event. Large language model based zero-shot classifier We leveraged the robust capabilities of LLMs as zero-shot annotators (Gilardi et al., 2023; Törnberg, 2023; Wang et al., 2023). LLMs have demonstrated their potential to be faster, more cost-effective, and, in many cases, more accurate than human text annotations. In particular, GPT has proven to be an effective tool for text annotation (Gilardi et al., 2023; Törnberg, 2023). As before, the quality of text annotations is heavily influenced by prompt engineering (Clavié et al., 2023; Sahoo et al., 2024). Therefore, we applied a range of best-practice prompting strategies (Clavié et al., 2023) to label farmer protest events (see Table 3 ). Additionally, we eliminated the variability in the generated responses by setting the temperature parameter to zero. Building on this foundation, we constructed a series of candidate classification models by varying several key factors: the prompt formulations (informed by elements that have proven to be most influential), namely the use of chain-of-thought (CoT) reasoning and the provision of label descriptions, as well as two distinct LLM models—GPT-4o and GPT-4o-mini. A summary of these candidate models is provided in Table 4 . Table 3 Overview of prompting strategies used for multilabel classification. Prompting strategy Description Implementation Role Role split Different roles are used for prompting. See rest of table below. System, User, Assistant Context provision Some context about the task and the role are provided. -You are [...] an intelligent expert at classifying textual information on farmer protests in Europe between 2023 and 2024. System Assign Name Assign a name to the model for reference in conversation. -You are Bob, an intelligent [...] System Chain of Thought Reasoning Reason step-by-step -Think step-by-step. System Multi-role task instruction The task instructions are designed as a combination of system and user instructions. -You are tasked with doing multiclass labeling on farmer protest descriptions. -[...] \n The LABELS are: \n {labels} -Analyze the following farmer protest description: \n\n {event_description} \n System, User Provision of label descriptions Further context is provided by presenting label desctiptions. -\n The following label descriptions apply: \n {labels}:{label_descriptions} System Strict output Instruct the model to respond strictly according to the provided template. - Your response MUST BE a single python list, no other text. -Provide the answer in the following format: LABELS. System, User Mock discussion Give task instructions by mocking a discussion where it acknowledges them -I will give you some text and you will analyze it. Got it? -Yes, I understand. I am Bob, and I will analyze the given task as best as I can. User, Assistant Provide positive feedback Offer the model positive feedback before presenting the query. -Great! Let's begin then :). User Table 4 LLM-based multilabel classifier variations. All prompting strategies presented in Table 3 apply, only chain-of thought and provision of descriptions vary. (-) not included (+) included. Model name Prompting variation GPT model Chain-of-Thought Provision of label descriptions nCoT_nCon_gpt-4o - - gpt-4o CoT_Con_gpt-4o + + gpt-4o nCoT_Con_gpt-4o - + gpt-4o CoT_nCon_gpt-4o + - gpt-4o nCoT_nCon_gpt-4-mini - - gpt-4o-mini CoT_Con_gpt-4o-mini + + gpt-4o-mini nCoT_Con_gpt-4o-mini - + gpt-4o-mini CoT_nCon_gpt-4o-mini + - gpt-4o-mini Human annotation and model selection To evaluate the performance of the candidate classification models, we randomly selected a subsample of event descriptions (n = 231) to serve as the evaluation set. Each LLM specification, as detailed in Table 4 , was tested on this subset. Additionally, we enlisted three research assistants with domain expertise in European agriculture and a background in agricultural science to classify the validation set. These annotators were provided with concise instructions (Supplementary information SI.3) to ensure consistency in their annotations. A key challenge in model selection is the absence of a ground truth for assigning labels to the text descriptions. Since no definitive "correct" labeling exists, it is difficult to directly determine which model or human annotator performs best. To address this issue, we employed an approach that selects the model most similar to human labeling. Specifically, we compared the outputs of all models and human annotations to assess their similarity, using this comparison as the basis for selecting the optimal model (Supplementary information SI.4). The selected final LLM annotation model was then applied to all protest instances to categorize each event based on the identified protest reasons. References Abay, K. A., Breisinger, C., Glauber, J., Kurdi, S., Laborde, D., & Siddig, K. (2023). The Russia-Ukraine war: Implications for global and regional food security and potential policy responses. Global Food Security , 36 , 100675. https://doi.org/10.1016/j.gfs.2023.100675 Abraham, A., Hassanien, A.-E., & Snášel, V. (Eds.). (2009). Foundations of Computational Intelligence Volume 5: Function Approximation and Classification (Vol. 205). Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-642-01536-6 ACLED. (2023). How does ACLED code and review data to ensure quality? https://acleddata.com/knowledge-base/how-does-acled-code-and-review-data-to-ensure-quality/ Anderson, W., Taylor, C., McDermid, S., Ilboudo-Nébié, E., Seager, R., Schlenker, W., Cottier, F., de Sherbinin, A., Mendeloff, D., & Markey, K. (2021). Violent conflict exacerbated drought-related food insecurity between 2009 and 2019 in sub-Saharan Africa. Nature Food , 2 (8), 603–615. https://doi.org/10.1038/s43016-021-00327-4 Aroyo, L., & Welty, C. (2015). Truth Is a Lie: Crowd Truth and the Seven Myths of Human Annotation. AI Magazine , 36 (1), 15–24. https://doi.org/10.1609/aimag.v36i1.2564 Ash, E., & Hansen, S. (2023). Text Algorithms in Economics. Annual Review of Economics , 15 (1), 659–688. https://doi.org/10.1146/annurev-economics-082222-074352 Behnassi, M., & El Haiba, M. (2022). Implications of the Russia–Ukraine war for global food security. Nature Human Behaviour , 6 (6), 754–755. https://doi.org/10.1038/s41562-022-01391-x Buitenhuis, Y., Candel, J. J. L., Termeer, K. J. A. M., & Feindt, P. H. (2020). Does the Common Agricultural Policy enhance farming systems’ resilience? Applying the Resilience Assessment Tool (ResAT) to a farming system case study in the Netherlands. Journal of Rural Studies , 80 , 314–327. https://doi.org/10.1016/j.jrurstud.2020.10.004 Burton, R. J. F., & Wilson, G. A. (2006). Injecting social psychology theory into conceptualisations of agricultural agency: Towards a post-productivist farmer self-identity? Journal of Rural Studies , 22 (1), 95–115. https://doi.org/10.1016/j.jrurstud.2005.07.004 Bush, E., & Simi, P. (2001). European Farmers and Their Protests. In Contentious Europeans: Protest and Politics in an Emerging Polity . Rowman & Littlefield. Chiang, W.-L., Zheng, L., Sheng, Y., Angelopoulos, A. N., Li, T., Li, D., Zhang, H., Zhu, B., Jordan, M., Gonzalez, J. E., & Stoica, I. (2024). Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference (No. arXiv:2403.04132). arXiv. https://doi.org/10.48550/arXiv.2403.04132 Clapp, J., & Moseley, W. G. (2020). This food crisis is different: COVID-19 and the fragility of the neoliberal food security order. The Journal of Peasant Studies , 47 (7), 1393–1417. https://doi.org/10.1080/03066150.2020.1823838 Clavié, B., Ciceu, A., Naylor, F., Soulié, G., & Brightwell, T. (2023). Large Language Models in the Workplace: A Case Study on Prompt Engineering for Job Type Classification (No. arXiv:2303.07142). arXiv. http://arxiv.org/abs/2303.07142 De Weerd, M., & Klandermans, B. (1999). Group identification and political protest: Farmers’ protest in the Netherlands. European Journal of Social Psychology , 29 (8), 1073–1095. https://doi.org/10.1002/(SICI)1099-0992(199912)29:83.0.CO;2-K Deininger, K., Ali, D. A., Kussul, N., Shelestov, A., Lemoine, G., & Yailimova, H. (2023). Quantifying war-induced crop losses in Ukraine in near real time to strengthen local and global food security. Food Policy , 115 , 102418. https://doi.org/10.1016/j.foodpol.2023.102418 Escritt, T. (2024). There’s no more money, German minister tells rowdy farmers. Reuters . https://www.reuters.com/world/europe/german-farmers-kick-off-protest-over-higher-taxes-berlin-2024-01-15/ European Parliament. (2023). Financing the CAP . https://www.europarl.europa.eu/factsheets/en/sheet/106/financing-of-the-cap. https://www.europarl.europa.eu/factsheets/en/sheet/106/financing-of-the-cap. Finger, R., Fabry, A., Kammer, M., Candel, J., Dalhaus, T., & Meemken, E. M. (2024). Farmer Protests in Europe 2023–2024. EuroChoices , 1746-692X.12452. https://doi.org/10.1111/1746-692X.12452 Foley, J. A., Ramankutty, N., Brauman, K. A., Cassidy, E. S., Gerber, J. S., Johnston, M., Mueller, N. D., O’Connell, C., Ray, D. K., West, P. C., Balzer, C., Bennett, E. M., Carpenter, S. R., Hill, J., Monfreda, C., Polasky, S., Rockström, J., Sheehan, J., Siebert, S., … Zaks, D. P. M. (2011). Solutions for a cultivated planet. Nature , 478 (7369), 337–342. https://doi.org/10.1038/nature10452 Forst, R. (2024, January 25). EU faces pressure to defuse mounting anger as farmers protest across Europe. Euronews . https://www.euronews.com/green/2024/01/25/eu-faces-pressure-to-defuse-mounting-anger-as-farmers-protest-across-europe Gaupp, F., Ruggeri Laderchi, C., Lotze-Campen, H., DeClerck, F., Bodirsky, B. L., Lowder, S., Popp, A., Kanbur, R., Edenhofer, O., Nugent, R., Fanzo, J., Dietz, S., Nordhagen, S., & Fan, S. (2021). Food system development pathways for healthy, nature-positive and inclusive food systems. Nature Food , 2 (12), 928–934. https://doi.org/10.1038/s43016-021-00421-7 Gilardi, F., Alizadeh, M., & Kubli, M. (2023). ChatGPT outperforms crowd workers for text-annotation tasks. Proceedings of the National Academy of Sciences of the United States of America , 120 (30), 1–3. https://doi.org/10.1073/pnas.2305016120 Ibrahim, H., Liu, F., Asim, R., Battu, B., Benabderrahmane, S., Alhafni, B., Adnan, W., Alhanai, T., AlShebli, B., Baghdadi, R., Bélanger, J. J., Beretta, E., Celik, K., Chaqfeh, M., Daqaq, M. F., Bernoussi, Z. E., Fougnie, D., Garcia de Soto, B., Gandolfi, A., … Zaki, Y. (2023). Perception, performance, and detectability of conversational artificial intelligence across 32 university courses. Scientific Reports , 13 (1), 12187. https://doi.org/10.1038/s41598-023-38964-3 Janker, J., Vesala, H. T., & Vesala, K. M. (2021). Exploring the link between farmers’ entrepreneurial identities and work wellbeing. Journal of Rural Studies , 83 , 117–126. https://doi.org/10.1016/j.jrurstud.2021.02.014 Macedonio Vega, R. (2024, November 19). Spanish farmers lash out against Mercosur: “They will end European agriculture.” Euronews . https://www.euronews.com/2024/11/19/spanish-farmers-lash-out-against-mercosur-they-will-end-european-agriculture Matthews, A. (2024a). Farmer Protests and Income Developments in the EU. The Political Quarterly , 95 (2), 344–349. https://doi.org/10.1111/1467-923X.13396 Matthews, A. (2024b). Farmer Protests and the 2024 European Parliament Elections. Intereconomics , 59 (2), 83–87. https://doi.org/10.2478/ie-2024-0018 Maystadt, J.-F., & Ecker, O. (2014). Extreme Weather and Civil War: Does Drought Fuel Conflict in Somalia through Livestock Price Shocks? American Journal of Agricultural Economics , 96 (4), 1157–1182. https://doi.org/10.1093/ajae/aau010 Mennig, P. (2024). A never-ending story of reforms: On the wicked nature of the Common Agricultural Policy. Npj Sustainable Agriculture , 2 (1), 20. https://doi.org/10.1038/s44264-024-00027-z Meuwissen, M. P. M., Feindt, P. H., Spiegel, A., Termeer, C. J. A. M., Mathijs, E., de Mey, Y., Finger, R., Balmann, A., Wauters, E., Urquhart, J., Vigani, M., Zawalińska, K., Herrera, H., Nicholas-Davies, P., Hansson, H., Paas, W., Slijper, T., Coopmans, I., Vroege, W., … Reidsma, P. (2019). A framework to assess the resilience of farming systems. Agricultural Systems , 176 (January), 102656. https://doi.org/10.1016/j.agsy.2019.102656 Miller, E., Kishi, R., Raleigh, C., & Dowd, C. (2022). An agenda for addressing bias in conflict data. Scientific Data , 9 (1), 593. https://doi.org/10.1038/s41597-022-01705-8 Mittelstädt, J. M., Maier, J., Goerke, P., Zinn, F., & Hermes, M. (2024). Large language models can outperform humans in social situational judgments. Scientific Reports , 14 (1), 27449. https://doi.org/10.1038/s41598-024-79048-0 Mustafa, Z., Vitali, G., Huffaker, R., & Canavari, M. (2024). A systematic review on price volatility in agriculture. Journal of Economic Surveys , 38 (1), 268–294. https://doi.org/10.1111/joes.12549 Naylor, E. L. (1994). Unionism, peasant protest and the reform of French agriculture. Journal of Rural Studies , 10 (3), 263–273. https://doi.org/10.1016/0743-0167(94)90053-1 O’Loughlin, J., Witmer, F. D. W., Linke, A. M., Laing, A., Gettelman, A., & Dudhia, J. (2012). Climate variability and conflict risk in East Africa, 1990–2009. Proceedings of the National Academy of Sciences , 109 (45), 18344–18349. https://doi.org/10.1073/pnas.1205130109 OpenAI, Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F. L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., Avila, R., Babuschkin, I., Balaji, S., Balcom, V., Baltescu, P., Bao, H., Bavarian, M., Belgum, J., … Zoph, B. (2023). GPT-4 Technical Report (Version 6). arXiv. https://doi.org/10.48550/ARXIV.2303.08774 Peeperkorn, M., Kouwenhoven, T., Brown, D., & Jordanous, A. (2024). Is Temperature the Creativity Parameter of Large Language Models? (No. arXiv:2405.00492). arXiv. https://doi.org/10.48550/arXiv.2405.00492 Pe’er, G., Dicks, L. V., Visconti, P., Arlettaz, R., Báldi, A., Benton, T. G., Collins, S., Dieterich, M., Gregory, R. D., Hartig, F., Henle, K., Hobson, P. R., Kleijn, D., Neumann, R. K., Robijns, T., Schmidt, J., Shwartz, A., Sutherland, W. J., Turbé, A., … Scott, A. V. (2014). EU agricultural reform fails on biodiversity. Science , 344 (6188), 1090–1092. https://doi.org/10.1126/science.1253425 Pe’er, G., Zinngrebe, Y., Moreira, F., Sirami, C., Schindler, S., Müller, R., Bontzorlos, V., Clough, D., Bezák, P., Bonn, A., Hansjürgens, B., Lomba, A., Möckel, S., Passoni, G., Schleyer, C., Schmidt, J., & Lakner, S. (2019). A greener path for the EU Common Agricultural Policy. Science , 365 (6452), 449–451. https://doi.org/10.1126/science.aax3146 Plümper, T., Neumayer, E., & Pfaff, K. G. (2021). The strategy of protest against Covid‐19 containment policies in Germany. Social Science Quarterly , 102 (5), 2236–2250. https://doi.org/10.1111/ssqu.13066 Raleigh, C., Kishi, R., & Linke, A. (2023). Political instability patterns are obscured by conflict dataset scope conditions, sources, and coding choices. Humanities and Social Sciences Communications , 10 (1), 74. https://doi.org/10.1057/s41599-023-01559-4 Raleigh, C., Linke, rew, Hegre, H., & Karlsen, J. (2010). Introducing ACLED: An Armed Conflict Location and Event Dataset. Journal of Peace Research , 47 (5), 651–660. https://doi.org/10.1177/0022343310378914 Sahoo, P., Singh, A. K., Saha, S., Jain, V., Mondal, S., & Chadha, A. (2024). A Systematic Survey of Prompt Engineering in Large Language Models: Techniques and Applications (No. arXiv:2402.07927). arXiv. http://arxiv.org/abs/2402.07927 Saleh, R., & Ehlers, M.-H. (2023). Exploring farmers’ perceptions of social sustainability. Environment, Development and Sustainability . https://doi.org/10.1007/s10668-023-04140-w Schebesta, H., & Candel, J. J. L. (2020). Game-changing potential of the EU’s Farm to Fork Strategy. Nature Food , 1 (10), 586–588. https://doi.org/10.1038/s43016-020-00166-9 Shuman, E., Hasan-Aslih, S., Van Zomeren, M., Saguy, T., & Halperin, E. (2022). Protest movements involving limited violence can sometimes be effective: Evidence from the 2020 BlackLivesMatter protests. Proceedings of the National Academy of Sciences , 119 (14), e2118990119. https://doi.org/10.1073/pnas.2118990119 Slijper, T., de Mey, Y., Poortvliet, P. M., & Meuwissen, M. P. M. (2022). Quantifying the resilience of European farms using FADN. European Review of Agricultural Economics , 49 (1), 121–150. https://doi.org/10.1093/erae/jbab042 Srivastava, A., Rastogi, A., Rao, A., Shoeb, A. A. M., Abid, A., Fisch, A., Brown, A. R., Santoro, A., Gupta, A., Garriga-Alonso, A., Kluska, A., Lewkowycz, A., Agarwal, A., Power, A., Ray, A., Warstadt, A., Kocurek, A. W., Safaya, A., Tazarv, A., … Wu, Z. (2022). Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models . https://doi.org/10.48550/ARXIV.2206.04615 Tilman, D., Balzer, C., Hill, J., & Befort, B. L. (2011). Global food demand and the sustainable intensification of agriculture. Proceedings of the National Academy of Sciences , 108 (50), 20260–20264. https://doi.org/10.1073/pnas.1116437108 Törnberg, P. (2023). ChatGPT-4 Outperforms Experts and Crowd Workers in Annotating Political Twitter Messages with Zero-Shot Learning (No. arXiv:2304.06588). arXiv. http://arxiv.org/abs/2304.06588 Wang, Y., Wang, W., Chen, Q., Huang, K., Nguyen, A., & De, S. (2023). Prompt-based Zero-shot Text Classification with Conceptual Knowledge. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) , 30–38. https://doi.org/10.18653/v1/2023.acl-srw.4 Webb, T., Holyoak, K. J., & Lu, H. (2023). Emergent analogical reasoning in large language models. Nature Human Behaviour , 7 (9), 1526–1541. https://doi.org/10.1038/s41562-023-01659-w Weidmann, N. B. (2016). A Closer Look at Reporting Bias in Conflict Event Data. American Journal of Political Science , 60 (1), 206–218. https://doi.org/10.1111/ajps.12196 Yang, Y., Tilman, D., Jin, Z., Smith, P., Barrett, C. B., Zhu, Y.-G., Burney, J., D’Odorico, P., Fantke, P., Fargione, J., Finlay, J. C., Rulli, M. C., Sloat, L., Jan van Groenigen, K., West, P. C., Ziska, L., Michalak, A. M., the Clim-Ag Team, Lobell, D. B., … Zhuang, M. (2024). Climate change exacerbates the environmental impacts of agriculture. Science , 385 (6713), eadn3747. https://doi.org/10.1126/science.adn3747 Zhao, W. X., Zhou, K., Li, J., Tang, T., Wang, X., Hou, Y., Min, Y., Zhang, B., Zhang, J., Dong, Z., Du, Y., Yang, C., Chen, Y., Chen, Z., Jiang, J., Ren, R., Li, Y., Tang, X., Liu, Z., … Wen, J.-R. (2024). A Survey of Large Language Models (No. arXiv:2303.18223). arXiv. https://doi.org/10.48550/arXiv.2303.18223 Footnotes The temperature parameter is a hyperparameter that influences the probability distribution used to select the next token in a sequence (Peeperkorn et al., 2024). A low temperature causes the model to focus on tokens with higher probabilities, resulting in more deterministic outputs. In contrast, a higher temperature increases the likelihood of selecting less probable tokens, which encourages more diversity and creativity. At a temperature of 1, the model samples directly from the raw softmax distribution without scaling, striking a balance between deterministic and stochastic behavior (Peeperkorn et al., 2024), which we found ideal for generating varied yet coherent responses. Additional Declarations There is NO Competing Interest. Supplementary Files supplementarymaterials.zip Supplementary Information codesanddata.zip Code and Data RS.pdf Reporting Summary Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6652927","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":472437642,"identity":"f2c80996-e2ed-4a88-9c95-d7a435f4e58b","order_by":0,"name":"Christian Stetter","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAt0lEQVRIie3PsQrCMBCA4YNAOma9vsVBoSCUPkvkoKMOLh27OfYRfAvnKwVdCq5dS8FZt2xacXRIBAeHfEuW+7kcQBT9MwOgX68NmFUAssyljfo2IQlNTHvp5rsrttm5PU0Oyo03wZEVia1Wx6FPCIF33oRQaRTbUz6yRgJZN97E9IkT+6DsMGu0QQmwXs4Xeq8LSZZbMhwqJhw4Txtif2LabrrVRUlm311TV5f+5OOfURRF0S88ATSTOgaqMhV7AAAAAElFTkSuQmCC","orcid":"","institution":"ETH Zürich","correspondingAuthor":true,"prefix":"","firstName":"Christian","middleName":"","lastName":"Stetter","suffix":""},{"id":472437643,"identity":"e623bc9a-795b-402e-a94f-7bba7cb692ba","order_by":1,"name":"Eva-Marie Meemken","email":"","orcid":"","institution":"Food System Economics and Policy Group, ETH Zurich","correspondingAuthor":false,"prefix":"","firstName":"Eva-Marie","middleName":"","lastName":"Meemken","suffix":""},{"id":472437644,"identity":"ea4a72df-678f-430c-8b36-89ca58a01284","order_by":2,"name":"Andrea Fürholz","email":"","orcid":"","institution":"Agricultural Economics and Policy Group, ETH Zürich","correspondingAuthor":false,"prefix":"","firstName":"Andrea","middleName":"","lastName":"Fürholz","suffix":""},{"id":472437645,"identity":"1f9666fa-c947-4d8d-bc3c-4b4e7f97c12c","order_by":3,"name":"Robert Finger","email":"","orcid":"https://orcid.org/0000-0002-0634-5742","institution":"ETH","correspondingAuthor":false,"prefix":"","firstName":"Robert","middleName":"","lastName":"Finger","suffix":""}],"badges":[],"createdAt":"2025-05-13 08:00:16","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6652927/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6652927/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":84868874,"identity":"c3a18b06-4fdc-4da2-bf3f-8a037a23066c","added_by":"auto","created_at":"2025-06-18 08:45:13","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1154787,"visible":true,"origin":"","legend":"Overview of main results of European farmer protest classification. The analysis is based on systematically processed textual descriptions of 4,303 farmer protests in Europe from November 1, 2023 to March 31, 2024, filtered from the Armed Conflict Location \u0026amp; Event Data Project (ACLED) database and analyzed using large language models (see Online methods). Protest reasons were identified, categorized, and validated using a combination of LLM-based categorization, zero-shot classification and human validation. Panel details: A) describes to how often the identified protest reasons were assigned to a protest event. B) maps the spatial distribution of protest events at large (central panel) and the most relevant protest reasons (surrounding figures). C) illustrates the interconnectedness of protest reason across Europe at large. Nodes reflect betweenness centrality, and edges reflect scaled co-occurrence of protest reasons. D) provides an overview of temporal evolution of protest reasons on a weekly basis. The y-axis represents the occurrence count of the reasons.","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6652927/v1/408979dabeafc0252d650f00.png"},{"id":84868880,"identity":"40c59112-fbf6-450c-823e-628dd8ad879d","added_by":"auto","created_at":"2025-06-18 08:45:13","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":568020,"visible":true,"origin":"","legend":"\u003cp\u003ePrevalence of farmers’ protest reasons across European countries with the highest farmer protest counts (≥35) from November 1, 2023 to March 31, 2024. The analysis is based on systematically processed textual descriptions of 4,303 farmer protests in Europe from November 1, 2023 to March 31, 2024, filtered from the Armed Conflict Location \u0026amp; Event Data Project (ACLED) database and analyzed using large language models (see Online methods). Protest reasons were identified, categorized, and validated using a combination of LLM-based categorization, zero-shot classification and human validation. Countries are listed according to the number of protests recorded (4,642).\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6652927/v1/d08c265c140057abca0137db.png"},{"id":84868885,"identity":"da0f3fdd-fe00-4c45-868a-e564f3adf72a","added_by":"auto","created_at":"2025-06-18 08:45:14","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":191528,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow for Categorizing and Classifying Protest Reasons. The methodology comprises two main components: the \"Proposer\" and the \"Multi-label Classifier.\" The Proposer begins with ACLED protest descriptions (November 1, 2023, to March 31, 2024), which are randomly sub-sampled and analyzed using GPT-4 to generate categories of protest reasons in 15 iterations. These suggestions are refined through human evaluation, resulting in a final list of protest categories and descriptions. The Multi-label Classifier uses these finalized labels to annotate protest events. Large language models (LLMs) are compared to human annotators to select the optimal (i.e. most human-like) LLM zero-shot annotation approach, which is then applied to the full ACLED dataset for classification\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6652927/v1/b5d576998de23899b236d229.png"},{"id":84869821,"identity":"e3a5f4b5-909e-4440-a696-afffdb7eb703","added_by":"auto","created_at":"2025-06-18 08:53:13","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":85540,"visible":true,"origin":"","legend":"\u003cp\u003eMethod of extracting data\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6652927/v1/0817051df321feff74bcdee4.png"},{"id":89062963,"identity":"047f4b1c-bfef-4bf2-a593-ae309c56fd9f","added_by":"auto","created_at":"2025-08-14 09:56:06","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2742596,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6652927/v1/2c323e48-53c0-4cb6-81e0-315c46ac25c3.pdf"},{"id":84868882,"identity":"3ab5fd6d-e05d-4708-9046-66a3a1feb80e","added_by":"auto","created_at":"2025-06-18 08:45:14","extension":"zip","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":1991476,"visible":true,"origin":"","legend":"Supplementary Information","description":"","filename":"supplementarymaterials.zip","url":"https://assets-eu.researchsquare.com/files/rs-6652927/v1/a427c62381950fee7e213856.zip"},{"id":84868887,"identity":"c9a41b2c-58a3-41a7-9ef5-e428393915e8","added_by":"auto","created_at":"2025-06-18 08:45:14","extension":"zip","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":42455221,"visible":true,"origin":"","legend":"Code and Data","description":"","filename":"codesanddata.zip","url":"https://assets-eu.researchsquare.com/files/rs-6652927/v1/f98071d791b163719c9dec43.zip"},{"id":84868875,"identity":"c5eacdcd-d3b8-481b-9d98-a111dac74ebe","added_by":"auto","created_at":"2025-06-18 08:45:13","extension":"pdf","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":104241,"visible":true,"origin":"","legend":"Reporting Summary","description":"","filename":"RS.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6652927/v1/4b6eb367a129d1600361d14e.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Large language model analysis reveals key reasons behind massive farmer protests in Europe","fulltext":[{"header":"Main text","content":" \u003cp\u003eThe agricultural sector, in Europe and worldwide, is facing unprecedented challenges: it must produce sufficient and diverse food for all, without exacerbating climate change and other environmental challenges, while ensuring farmers can earn a sustainable livelihood (Mustafa et al., 2024; Schebesta \u0026amp; Candel, 2020; Tilman et al., 2011; Yang et al., 2024). These issues have been further intensified by recent external shocks such as the COVID-19 pandemic and the war against Ukraine, which have disrupted global supply chains, amplified uncertainty, and increased the demand for greater food self-sufficiency (Abay et al., 2023; Behnassi \u0026amp; El Haiba, 2022; Clapp \u0026amp; Moseley, 2020). In Europe, farmers receive generous subsidies (e.g., 30% of the EU budget) designed to balance agricultural productivity, environmental protection, and farm profitability (European Parliament, 2023; Foley et al., 2011; Pe\u0026rsquo;er et al., 2019). Yet, farmers feel under immense pressure, believing that the support they receive is inadequate to address the challenges they face.\u003c/p\u003e \u003cp\u003eThis strain has culminated in massive farmer protests across Europe in 2023/24. Farmers took the streets to voice their frustration with the current situation and demanded action (Finger et al., 2024). Many farmers felt their livelihoods are at risk, struggling to balance environmentally sustainable practices with economic survival. Although farmer protests have a long-standing history in Europe and elsewhere (Bush \u0026amp; Simi, 2001; De Weerd \u0026amp; Klandermans, 1999; Naylor, 1994), the protests in 2023/24 were unique in their magnitude and geographical spread\u0026mdash;and have drawn widespread media attention. This led to prompt reactions from policy makers, significantly contributing to the reversal of ambitious environmental goals in European agriculture (Matthews, 2024b).\u003c/p\u003e \u003cp\u003eTo fully understand the dynamics of these protests and their potential impact on European agricultural policy, it is key to understand the underlying reasons of the protests and their development over time and space. The extant literature on the 2023/24 farmer protests is extremely limited. So far, research on these farmer protests have been based on anecdotal evidence and unsystematic evaluations of media reports (Finger et al., 2024; Matthews, 2024b, 2024a). These studies have highlighted pressure on farm incomes, tighter environmental regulations, and trade competition (Matthews, 2024a) as possible overarching drivers of farmer protests, which can differ across countries (Finger et al., 2024). However, systematic analyses of these protests are absent, and so is evidence on protest reasons across time and space.\u003c/p\u003e \u003cp\u003eThis study is the first to provide a comprehensive and systematic analysis of reasons and struggles driving the 2023/24 European farmer protests. We use data from the Armed Conflict Location \u0026amp; Event Data Project (ACLED), an extensive database capturing more than 4,600 farmer protest events across Europe from November 2023 to March 2024. The database contains location, date and qualitative information on individual protests and underlying themes and motivations (Raleigh et al., 2010, 2023). For our analysis, we leverage the capabilities of large language models to identify important protest reasons and classify farmer protest events accordingly (Ash \u0026amp; Hansen, 2023; Gilardi et al., 2023; Srivastava et al., 2022). Modern large language models (LLMs), such as GPT (OpenAI et al., 2023) have shown exceptional performance in various natural language processing tasks, including translation, question-answering and text classification, often surpassing human performance (Ibrahim et al., 2023; Mittelst\u0026auml;dt et al., 2024; T\u0026ouml;rnberg, 2023; Webb et al., 2023).We use LLMs to process the large volume of text data (i.e. qualitative descriptions of the 4,303 farmer protest events) to achieve two goals: i) to identify key reasons for farmer protests and ii) to classify protest events into categories aligned with these identified reasons. We apply this analysis over time, capturing the period from 1 November 2023 to 31 March 2024\u0026mdash;the main period of the protests\u0026mdash;and across space (i.e., covering all European countries). This allows us to uncover and quantify trends and patterns in the evolution of farmer protest movements, as well as regional or temporal variations.\u003c/p\u003e \u003cp\u003eA diverse set of reasons were behind the farmer protests\u003c/p\u003e \u003cp\u003eUsing the large corpus of available protest descriptions reported in the ACLED dataset, and employing an LLM-human collaborative approach (see Online methods), we detected 18 protest reasons that are described in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Reasons include a variety of interlinked and sometimes overlapping issues, such as (i) opposition to farm, trade, and environmental policies as well as (ii) dissatisfaction with market developments, prices, working conditions, and livelihood opportunities.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eLLM-human detected reasons for the European farmer protests from November 2023 - March 2024\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eProtest reasons\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDescriptions\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRising production costs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProtests driven by the increasing costs of production, including fuel, fertilizers, insurance, and energy, and other agricultural inputs, impacting farm profitability and contributing to financial pressures on farmers.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOpposition to EU free-trade agreements\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProtests and demonstrations against the ratification of the EU's free-trade agreements, demanding that partner countries comply with EU agricultural standards to ensure fair competition and protect local farmers' interests and maintain market stability.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEnvironmental regulations and agricultural standards\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProtests against various agricultural policies set by the European Union, including the Green Deal, and Common Agricultural Policy. Protesters oppose stricter environmental norms and regulations, particularly those related to pesticide and nitrogen use, which are perceived as overly restrictive and detrimental to agricultural practices.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBureaucratic constraints\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProtests against excessive bureaucratic processes and administrative burdens related to agricultural operations, state support, and subsidies that farmers believe hinder operational efficiency, productivity, sustainability in the agricultural sector. Protesters lament complex regulatory frameworks, excessive paperwork, and compliance requirements.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOpposition to foreign agricultural imports\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDemonstrations and actions targeting the import of agricultural products from foreign countries, particularly Ukraine, which are perceived to undercut local market prices and harm domestic farmers. The movements emphasize the need for fair production prices and express concerns over the quality of imported goods, which are seen as detrimental to consumer trust and local agricultural markets.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSubsidy cuts\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProtests against government plans to cut or delay subsidies and tax exemptions critical to the agricultural sector. These protests often focus on issues such as reductions in agricultural diesel subsidies, vehicle tax exemptions, and overall financial support necessary for farm operations. The protests reflect a broader demand for increased state support and highlight the economic challenges faced by the agricultural community due to inadequate subsidies and budget cuts.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClimate and natural disaster relief\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProtests and movements advocating for relief measures and compensation for farmers to offset the consequences of droughts, floods, and other climate-related natural disasters impacting agricultural production. These movements emphasize the urgent need for financial support, sustainable practices, and specific policies to mitigate the adverse impacts of climate change and environmental challenges on agricultural production.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLabor and social conditions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMovements and protests advocating for improved working and social conditions for farmers and farm workers, including disputes over working hours, fair pay, safety, financial security, health, and support measures during crises. They also call for greater recognition and appreciation of their work.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSolidarity movements\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDemonstrations showcasing support among farmers and other sectors, jointly advocating for farmers' recognition, and addressing shared economic challenges, regulatory burdens, and broader social issues such as anti-racism, immigrant rights, and social equality across Europe.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOpposition to renewable energy projects\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDemonstrations opposing large-scale renewable energy projects, such as wind and solar parks, due to concerns about their negative impact on agricultural land, practices, and local environments.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEconomic struggles and agricultural livelihoods\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProtests addressing the economic challenges faced by farmers, including low production prices, high costs, debt relief and demands for financial support, better working conditions and survival of the agricultural sector.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eConsumer awareness initiatives\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEfforts to raise awareness among consumers to educate the public on agricultural issues and the effects of policies on food production and pricing, often involving direct actions at retail locations.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLivestock and animal welfare protests\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProtests focused on the challenges faced by livestock breeders, including concerns over animal welfare regulations, insufficient financial support, market conditions, and the impact of EU policies on livestock and beekeeping, that are seen as detrimental to farmers' livelihoods.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOpposition to non-traditional products\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProtests against laboratory-made meat, insect-based flour, and other novel agricultural products perceived to threaten traditional farming practices and cultural authenticity.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInfrastructure and transport policies\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThis category encompasses various demonstrations aimed at addressing grievances related to infrastructure and trade policies. Key issues include unfavorable trade and transport policies affecting transport companies, particularly in cross-border scenarios, as well as local environmental concerns. The protests frequently involve stakeholders such as truck drivers, transport unions, and local communities advocating for fair labor regulations and sustainable infrastructure development.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNational and local state support\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProtests and demonstrations focused on the need for increased state financial support and subsidies for the agricultural sector, addressing issues like low compensation, economic difficulties, and specific local and national policies affecting farmers.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFair compensation and market practices\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProtests centered around low prices paid for agricultural products, which are seen as insufficient to cover the costs of production and ensure a viable income for farmers, protection for local farmers against low-cost foreign products, fairer compensation from large retailers, and addressing unfair trade practices that impact local farmers.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMiscellaneous agriculture-related protests\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProtests and demonstrations related to various miscellaneous aspects of the agricultural sector. It includes issues such as technological regulations, employment and training, public health and safety, immigration and trade, environmental sustainability, political demands, and unique local grievances.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"3\"\u003eNote:\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThese reasons were identified and categorized by leveraging GPT-4\u0026rsquo;s advanced natural language processing capabilities. By refining prompt engineering techniques, we generated a coherent classification of protest reasons, ensuring accuracy and contextual relevance. To address LLM limitations, we incorporated human refinement. These reasons and descriptions were provided to a GPT-4 to apply zero-shot multilabel classification to label each protest event during the European 2023/24 farmer protests according to the detected categories of reasons. By using state-of-the-art prompting strategies and validating the model against trained human annotators, we could ensure a coherent protest event classification. Details are provided in the Online methods and Supplementary Information (SI).\u003c/p\u003e \u003cp\u003eMajor driving forces and their spatial distribution across Europe\u003c/p\u003e \u003cp\u003eGeographically, farmer protests surged across Europe, with particularly high levels of protest activity, concentrated in Germany (1,072 protests or 23% of all protests), France (932, 20%), Poland (877, 19%), Spain (576, 12%), and Italy (266, 6%). These countries contributed to 81% of all protests. Scandinavian countries, the UK, and the Western Balkans saw much lower protest activity (accounting for less than 2% of all protests), underscoring regional disparities in the numbers of agricultural protests across the continent (see central panel in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb, and Supplementary information SI 6).\u003c/p\u003e \u003cp\u003eThe main results, illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, reveal the prevalence of the reasons for the protests (Panel A). Although protest reasons are strongly interconnected (for example, subsidy cuts often co-occurred with concerns over rising production costs and market conditions, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec and SI 6), there are different patterns across countries, underlined by looking at the following three most prevalent reasons.\u003c/p\u003e \u003cp\u003eFirst, opposition to foreign agricultural imports was the most common protest reason, having been associated with nearly 50% of protest events. This sentiment was especially strong in France, Poland, and Spain (see Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB and Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), where imports\u0026mdash;particularly from Ukraine\u0026mdash;were perceived as harmful to domestic farmers due to their competitive pricing. Many protests highlighted concerns about the perceived lower quality of imported goods and the increased market competition they brought (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe second most prevalent protest reason (~\u0026thinsp;30%) was opposition to various agricultural policies, including the Green Deal and Common Agricultural Policy of the European Union (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA), with particular emphasis in France, Italy, and Belgium (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Protesters expressed discontent with stricter environmental regulations, especially those limiting pesticide and nitrogen use, which were seen as overly restrictive and detrimental to agricultural productivity.\u003c/p\u003e \u003cp\u003eLastly, the third most common protest reason (~\u0026thinsp;27%) involved opposition to proposed government cuts or delays in subsidies and tax exemptions, notably in Germany (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC). These protests often centered around reductions in subsidies for agricultural diesel, vehicle tax exemptions, and general financial support to farm operations. The protests reflect a broader demand for increased state support and highlight the economic challenges faced by farmers despite policy support. Other significant factors included rising production costs, bureaucratic constraints, and dissatisfaction with market conditions.\u003c/p\u003e \u003cp\u003eThe prevalence of reasons varied not only across countries, but also over time (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ed), which is illustrated in more detail in the accompanying online application showing temporal and spatial heterogeneity: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://farmer-protest-app.onrender.com/\u003c/span\u003e\u003cspan address=\"https://farmer-protest-app.onrender.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e).\u003c/span\u003e Overall, we find that two initial waves were provoked by subsidy cuts, then reasons diversified into broader concerns about trade, policy, and economic struggles, and protests spread throughout much of Europe.\u003c/p\u003e "},{"header":"Discussion","content":"\u003cp\u003eAgriculture is facing unprecedented challenges that demand a transformative shift toward greater sustainability. In this context, large-scale farmer protests across Europe in 2023/24 have emerged as a powerful expression of widespread dissatisfaction, revealing the multifaceted struggles currently felt by many European farmers.\u003c/p\u003e \u003cp\u003eOur findings demonstrate that the reasons behind the European farmer protests were simultaneously heterogeneous, context-dependent, yet interconnected. The spatial distribution of the protests reveals important insights into the geography of farmers’ discontent. While France and the Netherlands have historically been focal points for farmer protests (De Weerd \u0026amp; Klandermans, 1999; Naylor, 1994), the farmer protests in 2023/24 marked a significant shift, characterized by high levels of protest events across the continent, and a contagious spread of these events in other parts of Europe, where such protests used to be less common.\u003c/p\u003e \u003cp\u003eThis shift suggests that farmer grievances have broadened geographically, indicating a growing transnational awareness of shared and mutual challenges. The geographical pattern of the protests reflects both the local specificity of farmers’ struggles and the growing interconnectedness of challenges European farmers are concerned with.\u003c/p\u003e \u003cp\u003eFor example, on the one hand, the European farmer protests of 2023/24 started locally in response to suggested cuts in agricultural diesel subsidies in Germany (Escritt, 2024), while in Poland, farmers’ discontent stemmed from rising production costs and fears over foreign agricultural imports, particularly from Ukraine (Finger et al., 2024). These differences highlight the importance of understanding local contexts, as farming systems across Europe operate under varying economic, social, and environmental conditions. Relatedly, not all regions exhibited the same level of farmer protests. Scandinavian countries, the United Kingdom, and parts of the Western Balkans reported relatively low protest activity. This unevenness raises questions about (i) whether some regions are better equipped to navigate agricultural challenges, which has in fact been found by Slijper et al. (2022) or (ii) if latent struggles remain unexpressed, e.g., due to weaker organizational capacity.\u003c/p\u003e \u003cp\u003eOn the other hand, the initial protests in Germany quickly gained traction in neighboring countries, as farmers recognized parallels in their own struggles. These developments illustrate how initial, localized struggles can escalate into transnational movements. This diffusion underscores how some of the contemporary struggles by farmers transcend national borders, fostering a shared sense of urgency among European farmers.\u003c/p\u003e \u003cp\u003eGenerally, the protest dynamics reveal a complex interplay of short-term shocks and long-term stresses. As Meuwissen et al. (2019) highlight, farming systems are shaped by these two categories of challenges (short-term shocks and long-term stresses). This dual framework demonstrates how short-term shocks can destabilize farming systems already weakened by long-term stresses, often leading to non-linear effects, heightened vulnerability, and an increased likelihood of tipping points. In the context of the protests, an acute policy shock—specifically the suggested Diesel subsidy cuts—triggered the initial demonstrations in Germany (Escritt, 2024). These quickly expanded to encompass a wide range of struggles reflecting (perceived) long-term stresses, such as administrative burden, red tape, rising production costs and competitive price levels. This evolution stresses why the protests became so widespread, exposing the deep frustrations that had been simmering beneath the surface long before the recent crises. Farmers view this eruption as an expression of the mounting pressures they have faced over time (Forst, 2024). Along these lines, Buitenhuis et al. (2020) show that the current common agricultural policy does not sufficiently support deeper levels of resilience within European agriculture.\u003c/p\u003e \u003cp\u003eFarmers in high-income countries have historically been well-organized and powerful. The 2023/24 protests represent a new form of farmers leveraging their power in public debates, utilizing media attention, and influencing policy decisions, such as those on environmental protection. Relatedly, protests have already erupted again in the winter of 2024/25, including France, Spain and Belgium, among other countries (Macedonio Vega, 2024), indicating this new quality of influence is here to stay. More broadly, these dynamics, therefore, suggest that structural resistance to change remains a significant barrier for transformation in European agrifood systems.\u003c/p\u003e \u003cp\u003eMany of the core reasons behind the protests are directly tied to farmers’ dissatisfaction with agricultural policies, and point to important goal conflicts between farm profitability, productivity, food self-sufficiency, and environmental protection – conflicts that these policies are designed to resolve but are not effectively solving in reality. Despite these existing tensions, farmer protests should not justify lowering environmental sustainability ambitions. While farmers perceive many regulations as burdensome and out of touch with their realities, policymakers face the challenge of balancing environmental imperatives with social and economic viability (Pe’er et al., 2019). Environmental regulations are essential to combat climate change and biodiversity loss (Pe’er et al., 2014, 2019), but the process of implementing such policies may have left many farmers feeling overwhelmed (Mennig, 2024).\u003c/p\u003e \u003cp\u003eUltimately, the protests highlight the need for a balanced approach to agricultural transformation. Previous research emphasizes the need to factor in the lived experiences of farmers when designing policies. Farmers’ self-concepts are often characterized by productivist and entrepreneurial identities and tied to their roles as stewards of the land (Burton \u0026amp; Wilson, 2006; Janker et al., 2021). Policy changes can threaten this self-concept or diminish their sense of purpose, not only leading to resistance but also threatening the social sustainability of European agriculture (Saleh \u0026amp; Ehlers, 2023). Future policy efforts must bridge the tensions between farmers’ subjective perceptions and objective realities by fostering stronger collaboration, communication, and coordination among policymakers and farmers (Mennig, 2024) — while also recognizing that transformations inevitably produce winners and losers, and thus offering new pathways or compensation for those whose livelihoods are threatened (Gaupp et al., 2021).\u003c/p\u003e \u003cp\u003eApart from that, this study demonstrates the value of large language models in the context of farmer protests and beyond. LLMs effectively classified farmer protest descriptions, offering new opportunities for understanding complex social phenomena.\u003c/p\u003e \u003cp\u003eAt the same time, this method has limitations. The data—short protest descriptions filtered by ACLED—capture only primary protest reasons, leaving secondary reasons, not reflected in the data, unexamined. Conflict and protest data, like that used by ACLED, often comes with inherent challenges related to reporting bias, which may stem from systematic inclinations, prejudices, and the directionality of information, influencing how these events are portrayed, how they are selected, and which aspects are emphasized (Miller et al., 2022; Weidmann, 2016). Although ACLED has strong quality control mechanisms (ACLED, 2023) that help reduce bias and maintain data reliability (see Online methods), it is impossible to completely eliminate bias from the process. Additionally, the LLM classifier cannot be tested against a ground truth, as the true nature of many events is inherently open to interpretation. This limitation, however, is not unique to machine-based classification; human annotators face the same challenge when classifying events, as their assessments are similarly influenced by context, perspective, and available information (Aroyo \u0026amp; Welty, 2015). To address this concern, we present evidence of agreement among human and machine annotators in the Supplementary Information.\u003c/p\u003e \u003cp\u003eFinally, future research could complement our analysis with in-depth interviews, surveys, and qualitative methods to better understand the underlying dynamics, identify the types of farmers involved in protests and capture secondary reasons, farmer demographics, and quantify protest impacts on outcomes such as policy responses, voting behavior or market prices.\u003c/p\u003e "},{"header":"Online Methods","content":"\u003cp\u003eOverview\u003c/p\u003e\u003cp\u003eThis section outlines the methodology used to analyze farmer protests in Europe from November 1, 2023 to March 31, 2024, using data from the Armed Conflict Location \u0026amp; Event Data Project (ACLED) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). To categorize the reasons behind these protests, a hybrid approach was employed, leveraging the capabilities of large language models (LLMs) with human refinement. This process involved systematically analyzing textual descriptions from ACLED, generating protest reason categories through GPT-4, and validating these categories via human evaluation. Subsequently, an LLM-based zero-shot multi-label classification model was developed to assign (multiple) protest reasons to individual events, enabling a comprehensive understanding of the underlying reasons and patterns of these demonstrations.\u003c/p\u003e\u003cp\u003eData description\u003c/p\u003e\u003cp\u003eThe data used in this study comes from the Armed Conflict Location \u0026amp; Event Data Project (ACLED), a comprehensive database that specializes in collecting and analyzing information on political violence, protests, and conflicts in various regions of the world (Raleigh et al., 2010). ACLED records events in real time, providing detailed data on incidents ranging from armed conflict and political violence to demonstrations and other forms of political unrest. This database is central to understand the local and regional dynamics of crises (Raleigh et al., 2010). The ACLED database contains over 1.5\u0026nbsp;million recorded events categorized into six primary types: combat, violence against civilians, remote violence (e.g., bombings), riots, protests, and strategic developments (Raleigh et al., 2023). Events often involve multiple actors, including governments, rebel groups, militias, or other organized groups with political goals or effects. It should be mentioned that events with and without violence are included (Raleigh et al., 2023). Relatedly, ACLED defines a protest event as a (peaceful) in-person demonstration by three or more persons (here farmers), at a specific date and location opposing a political, governmental, institutional, or private entity, where violence is absent on their part but may be directed against them (ACLED, 2024).\u003c/p\u003e\u003cp\u003eACLED's data collection follows a methodology that draws from a wide range of sources, including international, national, and local news agencies, nongovernmental organizations (NGOs), government reports, social media, and direct reports from eyewitnesses and regional partners. This multi-source approach ensures comprehensive coverage, especially in underreported or remote areas, and provides a more complete picture of global unrest. Details on the methodology are described in detail in (Raleigh et al., 2023), highlighting the importance of using diverse sources to capture localized conflict. In addition, ACLED's partnerships with more than 50 local conflict observatories and its reliance on more than 13,600 sources in more than 100 languages ensure a broad and detailed representation of events (Raleigh et al., 2023). Validation processes include inter-coder, intra-coder, inter-code, and inter-source reliability checks, as well as algorithmic checks and manual reviews by analysts to ensure data accuracy and reliability. ACLED emphasizes the credibility of sources and makes its methodology and sources publicly available to promote scientific accountability and allow for external verification (Raleigh et al., 2023).\u003c/p\u003e\u003cp\u003eThe database's applications have been demonstrated in numerous studies across various domains. For instance, Anderson et al. (2021) and Deininger et al. (2023) used data to analyze conflict and food insecurity, while Maystadt \u0026amp; Ecker (2014) and O’Loughlin et al. (2012) explored climate variability and conflict risk. Plümper et al. (2021) and Shuman et al. (2022) examined protest dynamics related to COVID-19 and the BlackLivesMatter movement using ACLED.\u003c/p\u003e\u003cp\u003eFor this analysis, the ACLED database was carefully filtered according to specific criteria to ensure the data's relevance to the research objectives. The selection process targeted events within the defined geographic region and timeframe, focusing on conflict types and actors pertinent to our study. Specifically, the data were restricted to events occurring between November 1, 2023, and March 31, 2024. Additionally, the filtering criteria required that protests take place in Europe, involving either farmers or organizations associated with them. The data extraction process is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e\u003cp\u003eClass construction and descriptions\u003c/p\u003e\u003cp\u003eTo distill the reasons behind these protests, it was necessary to systematically process the large corpus of text information provided. To that end, we leveraged the power of large language models (LLMs). LLMs have shown impressive performance in a wide range of tasks, including problem-solving and reasoning, natural language understanding and generation, as well as text summarization and classification, and even surpasses human performance in many of these tasks (Ibrahim et al., 2023; Mittelstädt et al., 2024; Törnberg, 2023; Webb et al., 2023). In terms of text analysis, a big benefit of LLMs compared to humans is that they can systematically analyze large corpuses of text in a short time at low cost (Gilardi et al., 2023).\u003c/p\u003e\u003cp\u003eTo construct classes of farmer protest reasons, we prompted a large language model (“Proposer”), namely GPT-4 (OpenAI et al., 2023), with the goal of obtaining a coherent list of classes that explain the reasons for the protests and their corresponding class descriptions. While there are several open and closed language models, OpenAI’s GPT has performed consistently well in recent years and GPT-4 is a benchmark in many natural language processing tasks (Chiang et al., 2024). GPT (Generative Pre-trained Transformer) is a neural network model built on the Transformer architecture, trained on vast datasets of text and code to understand and generate contextually relevant responses. It uses self-attention mechanisms to process sequential data efficiently (OpenAI et al., 2023). GPT, at its core, is a statistical model to generate responses. This means it uses probabilities to decide which token (a word or part of a word) to generate next. It predicts that next token by calculating the probabilities of all possible tokens based on the preceding context in the sequence (Zhao et al., 2024).\u003c/p\u003e\u003cp\u003eAs previous research on LLMs has demonstrated, the quality of their output is highly dependent on prompt engineering—specifically, how the user frames their task for the model (Sahoo et al., 2024). To tackle this, we explored multiple prompt variations through an iterative, interactive process, guided by established best practices in prompt engineering (Sahoo et al., 2024). We accessed GPT via OpenAI’s API, refining our approach by testing different formulations and selecting the one that produced the best results. The final prompt used to generate categories of protest reasons, and their corresponding descriptions is presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePrompting strategy for suggestion of reason categories.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"2\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSystem prompt\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYou are Bob, an intelligent expert at categorizing textual information on farmer protests in Europe between 2023 and 2024.\u003c/p\u003e \u003cp\u003eYou are tasked with discerning key issues, a.k.a. categories, from textual descriptions and reasons of farmer protests.\u003c/p\u003e \u003cp\u003eMake the categories diverse and distinguishable across, and coherent within.\u003c/p\u003e \u003cp\u003eOutput: You provide categories and a concise description for each category in table format only.\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUser 1 prompt\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eI will give you some text and you will analyze it. Got it?\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAssistant prompt\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes, I understand. I am Bob, and I will analyze the given task as best as I can.\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUser 2 prompt\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGreat! Let's begin then :). Analyze the following text: \\n\\n {text} \\n Answer: Let's reach a comprehensive categorization.\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e\u003cp\u003eTo address the inherent limitations of LLMs (Zhao et al., 2024), we incorporated human refinement to finalize the list of protest reasons and their corresponding descriptions. Initially, we divided the text corpus into three random subsets to ensure the content was manageable within GPT-4's token limits. For each subset, we ran five model iterations with the temperature hyperparameter set to one.[1]\u003ca class=\"FNLink\" href=\"#Fn1\" id=\"#FNLinkFn1\"\u003e\u003c/a\u003e\u003c/p\u003e\u003cp\u003eIn total, 15 responses were generated, each consisting of categories and corresponding descriptions (see Supplementary information SI.1). Two authors independently sorted these responses and identified patterns and coherences across them. For example, all 15 models identified a variation of the category \"Rising Production Costs.\" The sorted lists from both evaluators were subsequently compared (see Supplementary information SI.2). Based on this comparison, all authors reached a consensus on the final set of 18 labels and descriptions of the protest reasons.\u003c/p\u003e\u003cp\u003eIn a refinement step, the clarity and comprehensiveness of the proposed categories were evaluated by trained research assistants, who were tasked with labeling event descriptions according to the suggested protest reasons and descriptions. This process also served as a pre-test and training step for the subsequent classification task (see below). The feedback provided by the research assistants led to minor adjustments in the initially agreed-upon category descriptions to enhance clarity and avoid ambiguity.\u003c/p\u003e\u003cp\u003eMultilabel classification model\u003c/p\u003e\u003cp\u003eAfter establishing a list of potential protest reason categories in the initial step (“Proposer”), we employed multi-label classification to assign the relevant protest reasons to each protest event (“Multi-label Classifier”). Multi-label classification is a type of classification task where each instance can be associated with multiple classes, rather than just one. Unlike single-label classification such as binary and multiclass classification, where each instance is assigned only one class from a set of possible labels, multilabel classification allows for the prediction of multiple, simultaneous labels for a given instance (Abraham et al., 2009). This is the case in our application, where multiple reasons can simultaneously be associated with a single farmer protest event.\u003c/p\u003e\u003cp\u003eLarge language model based zero-shot classifier\u003c/p\u003e\u003cp\u003eWe leveraged the robust capabilities of LLMs as zero-shot annotators (Gilardi et al., 2023; Törnberg, 2023; Wang et al., 2023). LLMs have demonstrated their potential to be faster, more cost-effective, and, in many cases, more accurate than human text annotations. In particular, GPT has proven to be an effective tool for text annotation (Gilardi et al., 2023; Törnberg, 2023).\u003c/p\u003e\u003cp\u003eAs before, the quality of text annotations is heavily influenced by prompt engineering (Clavié et al., 2023; Sahoo et al., 2024). Therefore, we applied a range of best-practice prompting strategies (Clavié et al., 2023) to label farmer protest events (see Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Additionally, we eliminated the variability in the generated responses by setting the temperature parameter to zero.\u003c/p\u003e\u003cp\u003eBuilding on this foundation, we constructed a series of candidate classification models by varying several key factors: the prompt formulations (informed by elements that have proven to be most influential), namely the use of chain-of-thought (CoT) reasoning and the provision of label descriptions, as well as two distinct LLM models—GPT-4o and GPT-4o-mini. A summary of these candidate models is provided in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eOverview of prompting strategies used for multilabel classification.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrompting strategy\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDescription\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eImplementation\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRole\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRole split\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDifferent roles are used for prompting.\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cem\u003eSee rest of table below.\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSystem, User, Assistant\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eContext provision\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSome context about the task and the role are provided.\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-You are [...] an intelligent expert at classifying textual information on farmer protests in Europe between 2023 and 2024.\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSystem\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAssign Name\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAssign a name to the model for reference in conversation.\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-You are Bob, an intelligent [...]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSystem\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChain of Thought Reasoning\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eReason step-by-step\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-Think step-by-step.\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSystem\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMulti-role task instruction\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe task instructions are designed as a combination of system and user instructions.\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-You are tasked with doing multiclass labeling on farmer protest descriptions.\u003c/p\u003e \u003cp\u003e-[...] \\n The LABELS are: \\n {labels}\u003c/p\u003e \u003cp\u003e-Analyze the following farmer protest description: \\n\\n {event_description} \\n\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSystem, User\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProvision of label descriptions\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFurther context is provided by presenting label desctiptions.\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\\n The following label descriptions apply: \\n {labels}:{label_descriptions}\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSystem\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStrict output\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInstruct the model to respond strictly according to the provided template.\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Your response MUST BE a single python list, no other text.\u003c/p\u003e \u003cp\u003e-Provide the answer in the following format: LABELS.\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSystem, User\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMock discussion\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGive task instructions by mocking a discussion where it acknowledges them\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-I will give you some text and you will analyze it. Got it?\u003c/p\u003e \u003cp\u003e-Yes, I understand. I am Bob, and I will analyze the given task as best as I can.\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eUser, Assistant\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProvide positive feedback\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOffer the model positive feedback before presenting the query.\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-Great! Let's begin then :).\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eUser\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eLLM-based multilabel classifier variations. All prompting strategies presented in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e apply, only chain-of thought and provision of descriptions vary. (-) not included (+) included.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel name\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003ePrompting variation\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGPT model\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChain-of-Thought\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProvision of label descriptions\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003enCoT_nCon_gpt-4o\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003egpt-4o\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCoT_Con_gpt-4o\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003egpt-4o\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003enCoT_Con_gpt-4o\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003egpt-4o\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCoT_nCon_gpt-4o\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003egpt-4o\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003enCoT_nCon_gpt-4-mini\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003egpt-4o-mini\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCoT_Con_gpt-4o-mini\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003egpt-4o-mini\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003enCoT_Con_gpt-4o-mini\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003egpt-4o-mini\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCoT_nCon_gpt-4o-mini\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003egpt-4o-mini\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e\u003cp\u003eHuman annotation and model selection\u003c/p\u003e\u003cp\u003eTo evaluate the performance of the candidate classification models, we randomly selected a subsample of event descriptions (n = 231) to serve as the evaluation set. Each LLM specification, as detailed in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, was tested on this subset.\u003c/p\u003e\u003cp\u003eAdditionally, we enlisted three research assistants with domain expertise in European agriculture and a background in agricultural science to classify the validation set. These annotators were provided with concise instructions (Supplementary information SI.3) to ensure consistency in their annotations.\u003c/p\u003e\u003cp\u003eA key challenge in model selection is the absence of a ground truth for assigning labels to the text descriptions. Since no definitive \"correct\" labeling exists, it is difficult to directly determine which model or human annotator performs best. To address this issue, we employed an approach that selects the model most similar to human labeling. Specifically, we compared the outputs of all models and human annotations to assess their similarity, using this comparison as the basis for selecting the optimal model (Supplementary information SI.4). The selected final LLM annotation model was then applied to all protest instances to categorize each event based on the identified protest reasons.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAbay, K. A., Breisinger, C., Glauber, J., Kurdi, S., Laborde, D., \u0026amp; Siddig, K. (2023). The Russia-Ukraine war: Implications for global and regional food security and potential policy responses. \u003cem\u003eGlobal Food Security\u003c/em\u003e, \u003cem\u003e36\u003c/em\u003e, 100675. https://doi.org/10.1016/j.gfs.2023.100675\u003c/li\u003e\n\u003cli\u003eAbraham, A., Hassanien, A.-E., \u0026amp; Sn\u0026aacute;\u0026scaron;el, V. (Eds.). (2009). \u003cem\u003eFoundations of Computational Intelligence Volume 5: Function Approximation and Classification\u003c/em\u003e (Vol. 205). Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-642-01536-6\u003c/li\u003e\n\u003cli\u003eACLED. (2023). \u003cem\u003eHow does ACLED code and review data to ensure quality?\u003c/em\u003e https://acleddata.com/knowledge-base/how-does-acled-code-and-review-data-to-ensure-quality/\u003c/li\u003e\n\u003cli\u003eAnderson, W., Taylor, C., McDermid, S., Ilboudo-N\u0026eacute;bi\u0026eacute;, E., Seager, R., Schlenker, W., Cottier, F., de Sherbinin, A., Mendeloff, D., \u0026amp; Markey, K. (2021). Violent conflict exacerbated drought-related food insecurity between 2009 and 2019 in sub-Saharan Africa. \u003cem\u003eNature Food\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(8), 603\u0026ndash;615. https://doi.org/10.1038/s43016-021-00327-4\u003c/li\u003e\n\u003cli\u003eAroyo, L., \u0026amp; Welty, C. (2015). Truth Is a Lie: Crowd Truth and the Seven Myths of Human Annotation. \u003cem\u003eAI Magazine\u003c/em\u003e, \u003cem\u003e36\u003c/em\u003e(1), 15\u0026ndash;24. https://doi.org/10.1609/aimag.v36i1.2564\u003c/li\u003e\n\u003cli\u003eAsh, E., \u0026amp; Hansen, S. (2023). Text Algorithms in Economics. \u003cem\u003eAnnual Review of Economics\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e(1), 659\u0026ndash;688. https://doi.org/10.1146/annurev-economics-082222-074352\u003c/li\u003e\n\u003cli\u003eBehnassi, M., \u0026amp; El Haiba, M. (2022). Implications of the Russia\u0026ndash;Ukraine war for global food security. \u003cem\u003eNature Human Behaviour\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e(6), 754\u0026ndash;755. https://doi.org/10.1038/s41562-022-01391-x\u003c/li\u003e\n\u003cli\u003eBuitenhuis, Y., Candel, J. J. L., Termeer, K. J. A. M., \u0026amp; Feindt, P. H. (2020). Does the Common Agricultural Policy enhance farming systems\u0026rsquo; resilience? Applying the Resilience Assessment Tool (ResAT) to a farming system case study in the Netherlands. \u003cem\u003eJournal of Rural Studies\u003c/em\u003e, \u003cem\u003e80\u003c/em\u003e, 314\u0026ndash;327. https://doi.org/10.1016/j.jrurstud.2020.10.004\u003c/li\u003e\n\u003cli\u003eBurton, R. J. F., \u0026amp; Wilson, G. A. (2006). Injecting social psychology theory into conceptualisations of agricultural agency: Towards a post-productivist farmer self-identity? \u003cem\u003eJournal of Rural Studies\u003c/em\u003e, \u003cem\u003e22\u003c/em\u003e(1), 95\u0026ndash;115. https://doi.org/10.1016/j.jrurstud.2005.07.004\u003c/li\u003e\n\u003cli\u003eBush, E., \u0026amp; Simi, P. (2001). European Farmers and Their Protests. In \u003cem\u003eContentious Europeans: Protest and Politics in an Emerging Polity\u003c/em\u003e. Rowman \u0026amp; Littlefield.\u003c/li\u003e\n\u003cli\u003eChiang, W.-L., Zheng, L., Sheng, Y., Angelopoulos, A. N., Li, T., Li, D., Zhang, H., Zhu, B., Jordan, M., Gonzalez, J. E., \u0026amp; Stoica, I. (2024). \u003cem\u003eChatbot Arena: An Open Platform for Evaluating LLMs by Human Preference\u003c/em\u003e (No. arXiv:2403.04132). arXiv. https://doi.org/10.48550/arXiv.2403.04132\u003c/li\u003e\n\u003cli\u003eClapp, J., \u0026amp; Moseley, W. G. (2020). This food crisis is different: COVID-19 and the fragility of the neoliberal food security order. \u003cem\u003eThe Journal of Peasant Studies\u003c/em\u003e, \u003cem\u003e47\u003c/em\u003e(7), 1393\u0026ndash;1417. https://doi.org/10.1080/03066150.2020.1823838\u003c/li\u003e\n\u003cli\u003eClavi\u0026eacute;, B., Ciceu, A., Naylor, F., Souli\u0026eacute;, G., \u0026amp; Brightwell, T. (2023). \u003cem\u003eLarge Language Models in the Workplace: A Case Study on Prompt Engineering for Job Type Classification\u003c/em\u003e (No. arXiv:2303.07142). arXiv. http://arxiv.org/abs/2303.07142\u003c/li\u003e\n\u003cli\u003eDe Weerd, M., \u0026amp; Klandermans, B. (1999). Group identification and political protest: Farmers\u0026rsquo; protest in the Netherlands. \u003cem\u003eEuropean Journal of Social Psychology\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e(8), 1073\u0026ndash;1095. https://doi.org/10.1002/(SICI)1099-0992(199912)29:8\u0026lt;1073::AID-EJSP986\u0026gt;3.0.CO;2-K\u003c/li\u003e\n\u003cli\u003eDeininger, K., Ali, D. A., Kussul, N., Shelestov, A., Lemoine, G., \u0026amp; Yailimova, H. (2023). Quantifying war-induced crop losses in Ukraine in near real time to strengthen local and global food security. \u003cem\u003eFood Policy\u003c/em\u003e, \u003cem\u003e115\u003c/em\u003e, 102418. https://doi.org/10.1016/j.foodpol.2023.102418\u003c/li\u003e\n\u003cli\u003eEscritt, T. (2024). There\u0026rsquo;s no more money, German minister tells rowdy farmers. \u003cem\u003eReuters\u003c/em\u003e. https://www.reuters.com/world/europe/german-farmers-kick-off-protest-over-higher-taxes-berlin-2024-01-15/\u003c/li\u003e\n\u003cli\u003eEuropean Parliament. (2023). \u003cem\u003eFinancing the CAP\u003c/em\u003e. https://www.europarl.europa.eu/factsheets/en/sheet/106/financing-of-the-cap. https://www.europarl.europa.eu/factsheets/en/sheet/106/financing-of-the-cap.\u003c/li\u003e\n\u003cli\u003eFinger, R., Fabry, A., Kammer, M., Candel, J., Dalhaus, T., \u0026amp; Meemken, E. M. (2024). Farmer Protests in Europe 2023\u0026ndash;2024. \u003cem\u003eEuroChoices\u003c/em\u003e, 1746-692X.12452. https://doi.org/10.1111/1746-692X.12452\u003c/li\u003e\n\u003cli\u003eFoley, J. A., Ramankutty, N., Brauman, K. A., Cassidy, E. S., Gerber, J. S., Johnston, M., Mueller, N. D., O\u0026rsquo;Connell, C., Ray, D. K., West, P. C., Balzer, C., Bennett, E. M., Carpenter, S. R., Hill, J., Monfreda, C., Polasky, S., Rockstr\u0026ouml;m, J., Sheehan, J., Siebert, S., \u0026hellip; Zaks, D. P. M. (2011). Solutions for a cultivated planet. \u003cem\u003eNature\u003c/em\u003e, \u003cem\u003e478\u003c/em\u003e(7369), 337\u0026ndash;342. https://doi.org/10.1038/nature10452\u003c/li\u003e\n\u003cli\u003eForst, R. (2024, January 25). EU faces pressure to defuse mounting anger as farmers protest across Europe. \u003cem\u003eEuronews\u003c/em\u003e. https://www.euronews.com/green/2024/01/25/eu-faces-pressure-to-defuse-mounting-anger-as-farmers-protest-across-europe\u003c/li\u003e\n\u003cli\u003eGaupp, F., Ruggeri Laderchi, C., Lotze-Campen, H., DeClerck, F., Bodirsky, B. L., Lowder, S., Popp, A., Kanbur, R., Edenhofer, O., Nugent, R., Fanzo, J., Dietz, S., Nordhagen, S., \u0026amp; Fan, S. (2021). Food system development pathways for healthy, nature-positive and inclusive food systems. \u003cem\u003eNature Food\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(12), 928\u0026ndash;934. https://doi.org/10.1038/s43016-021-00421-7\u003c/li\u003e\n\u003cli\u003eGilardi, F., Alizadeh, M., \u0026amp; Kubli, M. (2023). ChatGPT outperforms crowd workers for text-annotation tasks. \u003cem\u003eProceedings of the National Academy of Sciences of the United States of America\u003c/em\u003e, \u003cem\u003e120\u003c/em\u003e(30), 1\u0026ndash;3. https://doi.org/10.1073/pnas.2305016120\u003c/li\u003e\n\u003cli\u003eIbrahim, H., Liu, F., Asim, R., Battu, B., Benabderrahmane, S., Alhafni, B., Adnan, W., Alhanai, T., AlShebli, B., Baghdadi, R., B\u0026eacute;langer, J. J., Beretta, E., Celik, K., Chaqfeh, M., Daqaq, M. F., Bernoussi, Z. E., Fougnie, D., Garcia de Soto, B., Gandolfi, A., \u0026hellip; Zaki, Y. (2023). Perception, performance, and detectability of conversational artificial intelligence across 32 university courses. \u003cem\u003eScientific Reports\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(1), 12187. https://doi.org/10.1038/s41598-023-38964-3\u003c/li\u003e\n\u003cli\u003eJanker, J., Vesala, H. T., \u0026amp; Vesala, K. M. (2021). Exploring the link between farmers\u0026rsquo; entrepreneurial identities and work wellbeing. \u003cem\u003eJournal of Rural Studies\u003c/em\u003e, \u003cem\u003e83\u003c/em\u003e, 117\u0026ndash;126. https://doi.org/10.1016/j.jrurstud.2021.02.014\u003c/li\u003e\n\u003cli\u003eMacedonio Vega, R. (2024, November 19). Spanish farmers lash out against Mercosur: \u0026ldquo;They will end European agriculture.\u0026rdquo; \u003cem\u003eEuronews\u003c/em\u003e. https://www.euronews.com/2024/11/19/spanish-farmers-lash-out-against-mercosur-they-will-end-european-agriculture\u003c/li\u003e\n\u003cli\u003eMatthews, A. (2024a). Farmer Protests and Income Developments in the EU. \u003cem\u003eThe Political Quarterly\u003c/em\u003e, \u003cem\u003e95\u003c/em\u003e(2), 344\u0026ndash;349. https://doi.org/10.1111/1467-923X.13396\u003c/li\u003e\n\u003cli\u003eMatthews, A. (2024b). Farmer Protests and the 2024 European Parliament Elections. \u003cem\u003eIntereconomics\u003c/em\u003e, \u003cem\u003e59\u003c/em\u003e(2), 83\u0026ndash;87. https://doi.org/10.2478/ie-2024-0018\u003c/li\u003e\n\u003cli\u003eMaystadt, J.-F., \u0026amp; Ecker, O. (2014). Extreme Weather and Civil War: Does Drought Fuel Conflict in Somalia through Livestock Price Shocks? \u003cem\u003eAmerican Journal of Agricultural Economics\u003c/em\u003e, \u003cem\u003e96\u003c/em\u003e(4), 1157\u0026ndash;1182. https://doi.org/10.1093/ajae/aau010\u003c/li\u003e\n\u003cli\u003eMennig, P. (2024). A never-ending story of reforms: On the wicked nature of the Common Agricultural Policy. \u003cem\u003eNpj Sustainable Agriculture\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(1), 20. https://doi.org/10.1038/s44264-024-00027-z\u003c/li\u003e\n\u003cli\u003eMeuwissen, M. P. M., Feindt, P. H., Spiegel, A., Termeer, C. J. A. M., Mathijs, E., de Mey, Y., Finger, R., Balmann, A., Wauters, E., Urquhart, J., Vigani, M., Zawalińska, K., Herrera, H., Nicholas-Davies, P., Hansson, H., Paas, W., Slijper, T., Coopmans, I., Vroege, W., \u0026hellip; Reidsma, P. (2019). A framework to assess the resilience of farming systems. \u003cem\u003eAgricultural Systems\u003c/em\u003e, \u003cem\u003e176\u003c/em\u003e(January), 102656. https://doi.org/10.1016/j.agsy.2019.102656\u003c/li\u003e\n\u003cli\u003eMiller, E., Kishi, R., Raleigh, C., \u0026amp; Dowd, C. (2022). An agenda for addressing bias in conflict data. \u003cem\u003eScientific Data\u003c/em\u003e, \u003cem\u003e9\u003c/em\u003e(1), 593. https://doi.org/10.1038/s41597-022-01705-8\u003c/li\u003e\n\u003cli\u003eMittelst\u0026auml;dt, J. M., Maier, J., Goerke, P., Zinn, F., \u0026amp; Hermes, M. (2024). Large language models can outperform humans in social situational judgments. \u003cem\u003eScientific Reports\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(1), 27449. https://doi.org/10.1038/s41598-024-79048-0\u003c/li\u003e\n\u003cli\u003eMustafa, Z., Vitali, G., Huffaker, R., \u0026amp; Canavari, M. (2024). A systematic review on price volatility in agriculture. \u003cem\u003eJournal of Economic Surveys\u003c/em\u003e, \u003cem\u003e38\u003c/em\u003e(1), 268\u0026ndash;294. https://doi.org/10.1111/joes.12549\u003c/li\u003e\n\u003cli\u003eNaylor, E. L. (1994). Unionism, peasant protest and the reform of French agriculture. \u003cem\u003eJournal of Rural Studies\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(3), 263\u0026ndash;273. https://doi.org/10.1016/0743-0167(94)90053-1\u003c/li\u003e\n\u003cli\u003eO\u0026rsquo;Loughlin, J., Witmer, F. D. W., Linke, A. M., Laing, A., Gettelman, A., \u0026amp; Dudhia, J. (2012). Climate variability and conflict risk in East Africa, 1990\u0026ndash;2009. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e, \u003cem\u003e109\u003c/em\u003e(45), 18344\u0026ndash;18349. https://doi.org/10.1073/pnas.1205130109\u003c/li\u003e\n\u003cli\u003eOpenAI, Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F. L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., Avila, R., Babuschkin, I., Balaji, S., Balcom, V., Baltescu, P., Bao, H., Bavarian, M., Belgum, J., \u0026hellip; Zoph, B. (2023). \u003cem\u003eGPT-4 Technical Report\u003c/em\u003e (Version 6). arXiv. https://doi.org/10.48550/ARXIV.2303.08774\u003c/li\u003e\n\u003cli\u003ePeeperkorn, M., Kouwenhoven, T., Brown, D., \u0026amp; Jordanous, A. (2024). \u003cem\u003eIs Temperature the Creativity Parameter of Large Language Models?\u003c/em\u003e (No. arXiv:2405.00492). arXiv. https://doi.org/10.48550/arXiv.2405.00492\u003c/li\u003e\n\u003cli\u003ePe\u0026rsquo;er, G., Dicks, L. V., Visconti, P., Arlettaz, R., B\u0026aacute;ldi, A., Benton, T. G., Collins, S., Dieterich, M., Gregory, R. D., Hartig, F., Henle, K., Hobson, P. R., Kleijn, D., Neumann, R. K., Robijns, T., Schmidt, J., Shwartz, A., Sutherland, W. J., Turb\u0026eacute;, A., \u0026hellip; Scott, A. V. (2014). EU agricultural reform fails on biodiversity. \u003cem\u003eScience\u003c/em\u003e, \u003cem\u003e344\u003c/em\u003e(6188), 1090\u0026ndash;1092. https://doi.org/10.1126/science.1253425\u003c/li\u003e\n\u003cli\u003ePe\u0026rsquo;er, G., Zinngrebe, Y., Moreira, F., Sirami, C., Schindler, S., M\u0026uuml;ller, R., Bontzorlos, V., Clough, D., Bez\u0026aacute;k, P., Bonn, A., Hansj\u0026uuml;rgens, B., Lomba, A., M\u0026ouml;ckel, S., Passoni, G., Schleyer, C., Schmidt, J., \u0026amp; Lakner, S. (2019). A greener path for the EU Common Agricultural Policy. \u003cem\u003eScience\u003c/em\u003e, \u003cem\u003e365\u003c/em\u003e(6452), 449\u0026ndash;451. https://doi.org/10.1126/science.aax3146\u003c/li\u003e\n\u003cli\u003ePl\u0026uuml;mper, T., Neumayer, E., \u0026amp; Pfaff, K. G. (2021). The strategy of protest against Covid‐19 containment policies in Germany. \u003cem\u003eSocial Science Quarterly\u003c/em\u003e, \u003cem\u003e102\u003c/em\u003e(5), 2236\u0026ndash;2250. https://doi.org/10.1111/ssqu.13066\u003c/li\u003e\n\u003cli\u003eRaleigh, C., Kishi, R., \u0026amp; Linke, A. (2023). Political instability patterns are obscured by conflict dataset scope conditions, sources, and coding choices. \u003cem\u003eHumanities and Social Sciences Communications\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(1), 74. https://doi.org/10.1057/s41599-023-01559-4\u003c/li\u003e\n\u003cli\u003eRaleigh, C., Linke, rew, Hegre, H., \u0026amp; Karlsen, J. (2010). Introducing ACLED: An Armed Conflict Location and Event Dataset. \u003cem\u003eJournal of Peace Research\u003c/em\u003e, \u003cem\u003e47\u003c/em\u003e(5), 651\u0026ndash;660. https://doi.org/10.1177/0022343310378914\u003c/li\u003e\n\u003cli\u003eSahoo, P., Singh, A. K., Saha, S., Jain, V., Mondal, S., \u0026amp; Chadha, A. (2024). \u003cem\u003eA Systematic Survey of Prompt Engineering in Large Language Models: Techniques and Applications\u003c/em\u003e (No. arXiv:2402.07927). arXiv. http://arxiv.org/abs/2402.07927\u003c/li\u003e\n\u003cli\u003eSaleh, R., \u0026amp; Ehlers, M.-H. (2023). Exploring farmers\u0026rsquo; perceptions of social sustainability. \u003cem\u003eEnvironment, Development and Sustainability\u003c/em\u003e. https://doi.org/10.1007/s10668-023-04140-w\u003c/li\u003e\n\u003cli\u003eSchebesta, H., \u0026amp; Candel, J. J. L. (2020). Game-changing potential of the EU\u0026rsquo;s Farm to Fork Strategy. \u003cem\u003eNature Food\u003c/em\u003e, \u003cem\u003e1\u003c/em\u003e(10), 586\u0026ndash;588. https://doi.org/10.1038/s43016-020-00166-9\u003c/li\u003e\n\u003cli\u003eShuman, E., Hasan-Aslih, S., Van Zomeren, M., Saguy, T., \u0026amp; Halperin, E. (2022). Protest movements involving limited violence can sometimes be effective: Evidence from the 2020 BlackLivesMatter protests. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e, \u003cem\u003e119\u003c/em\u003e(14), e2118990119. https://doi.org/10.1073/pnas.2118990119\u003c/li\u003e\n\u003cli\u003eSlijper, T., de Mey, Y., Poortvliet, P. M., \u0026amp; Meuwissen, M. P. M. (2022). Quantifying the resilience of European farms using FADN. \u003cem\u003eEuropean Review of Agricultural Economics\u003c/em\u003e, \u003cem\u003e49\u003c/em\u003e(1), 121\u0026ndash;150. https://doi.org/10.1093/erae/jbab042\u003c/li\u003e\n\u003cli\u003eSrivastava, A., Rastogi, A., Rao, A., Shoeb, A. A. M., Abid, A., Fisch, A., Brown, A. R., Santoro, A., Gupta, A., Garriga-Alonso, A., Kluska, A., Lewkowycz, A., Agarwal, A., Power, A., Ray, A., Warstadt, A., Kocurek, A. W., Safaya, A., Tazarv, A., \u0026hellip; Wu, Z. (2022). \u003cem\u003eBeyond the Imitation Game: Quantifying and extrapolating the capabilities of language models\u003c/em\u003e. https://doi.org/10.48550/ARXIV.2206.04615\u003c/li\u003e\n\u003cli\u003eTilman, D., Balzer, C., Hill, J., \u0026amp; Befort, B. L. (2011). Global food demand and the sustainable intensification of agriculture. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e, \u003cem\u003e108\u003c/em\u003e(50), 20260\u0026ndash;20264. https://doi.org/10.1073/pnas.1116437108\u003c/li\u003e\n\u003cli\u003eT\u0026ouml;rnberg, P. (2023). \u003cem\u003eChatGPT-4 Outperforms Experts and Crowd Workers in Annotating Political Twitter Messages with Zero-Shot Learning\u003c/em\u003e (No. arXiv:2304.06588). arXiv. http://arxiv.org/abs/2304.06588\u003c/li\u003e\n\u003cli\u003eWang, Y., Wang, W., Chen, Q., Huang, K., Nguyen, A., \u0026amp; De, S. (2023). Prompt-based Zero-shot Text Classification with Conceptual Knowledge. \u003cem\u003eProceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop)\u003c/em\u003e, 30\u0026ndash;38. https://doi.org/10.18653/v1/2023.acl-srw.4\u003c/li\u003e\n\u003cli\u003eWebb, T., Holyoak, K. J., \u0026amp; Lu, H. (2023). Emergent analogical reasoning in large language models. \u003cem\u003eNature Human Behaviour\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e(9), 1526\u0026ndash;1541. https://doi.org/10.1038/s41562-023-01659-w\u003c/li\u003e\n\u003cli\u003eWeidmann, N. B. (2016). A Closer Look at Reporting Bias in Conflict Event Data. \u003cem\u003eAmerican Journal of Political Science\u003c/em\u003e, \u003cem\u003e60\u003c/em\u003e(1), 206\u0026ndash;218. https://doi.org/10.1111/ajps.12196\u003c/li\u003e\n\u003cli\u003eYang, Y., Tilman, D., Jin, Z., Smith, P., Barrett, C. B., Zhu, Y.-G., Burney, J., D\u0026rsquo;Odorico, P., Fantke, P., Fargione, J., Finlay, J. C., Rulli, M. C., Sloat, L., Jan van Groenigen, K., West, P. C., Ziska, L., Michalak, A. M., the Clim-Ag Team, Lobell, D. B., \u0026hellip; Zhuang, M. (2024). Climate change exacerbates the environmental impacts of agriculture. \u003cem\u003eScience\u003c/em\u003e, \u003cem\u003e385\u003c/em\u003e(6713), eadn3747. https://doi.org/10.1126/science.adn3747\u003c/li\u003e\n\u003cli\u003eZhao, W. X., Zhou, K., Li, J., Tang, T., Wang, X., Hou, Y., Min, Y., Zhang, B., Zhang, J., Dong, Z., Du, Y., Yang, C., Chen, Y., Chen, Z., Jiang, J., Ren, R., Li, Y., Tang, X., Liu, Z., \u0026hellip; Wen, J.-R. (2024). \u003cem\u003eA Survey of Large Language Models\u003c/em\u003e (No. arXiv:2303.18223). arXiv. https://doi.org/10.48550/arXiv.2303.18223\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Footnotes","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003e The temperature parameter is a hyperparameter that influences the probability distribution used to select the next token in a sequence (Peeperkorn et al., 2024). A low temperature causes the model to focus on tokens with higher probabilities, resulting in more deterministic outputs. In contrast, a higher temperature increases the likelihood of selecting less probable tokens, which encourages more diversity and creativity. At a temperature of 1, the model samples directly from the raw softmax distribution without scaling, striking a balance between deterministic and stochastic behavior (Peeperkorn et al., 2024), which we found ideal for generating varied yet coherent responses.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6652927/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6652927/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"European farmers face mounting pressures. They must produce food while navigating climate change and preserving ecosystems. In 2023/24, these tensions erupted into massive protests across the continent. Despite their importance, farmer protests in Europe have received little attention and its drivers remain largely unclear. We address this gap by using large language models to analyze comprehensive data on 4,642 farmer protests throughout Europe from November 2023 to March 2024, and assess the main reasons for these protests. We find a complex interplay of economic, political and social pressures. Key reasons include opposition to foreign imports, resistance to EU policies, and frustration over subsidy cuts and delays. The analysis highlights significant geographical and temporal variations, with major protest hotspots in France, Germany, Spain, and Poland. Our findings highlight the need for solutions to balance environmental sustainability with profitability and ensuring a just transition for all stakeholders in the agrifood system.","manuscriptTitle":"Large language model analysis reveals key reasons behind massive farmer protests in Europe","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-18 08:45:09","doi":"10.21203/rs.3.rs-6652927/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"710759ab-573e-45f3-98ed-85b093c09777","owner":[],"postedDate":"June 18th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":50169416,"name":"Scientific community and society/Agriculture"},{"id":50169417,"name":"Scientific community and society/Social sciences/Interdisciplinary studies"}],"tags":[],"updatedAt":"2025-08-07T08:05:33+00:00","versionOfRecord":[],"versionCreatedAt":"2025-06-18 08:45:09","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6652927","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6652927","identity":"rs-6652927","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}