Extracting and analyzing urban housing conflicts using large language models,graph databases,and GIS

Abstract

Urban housing conflicts are increasingly shaping the social and spatial dynamics of cities, yet they remain difficult to analyze systematically due to their multi-actor complexity and fragmented representation across textual and spatial data. Existing studies tend to focus on either narrative or spatial aspects, rarely capturing the structural, temporal, and geographic dimensions of these conflicts in an integrated way. This paper addresses this gap by proposing a reproducible methodological framework that combines Large Language Models (LLMs), graph databases, and Geographic Information Systems (GIS) to analyze housing conflicts in Montréal between 2001 and 2024. The study aims to demonstrate how urban housing conflicts can be systematically extracted, classified, and analyzed across time and space using AI-based methods, and how their structural patterns reflect underlying socio-political dynamics. The resulting framework offers new insights into the evolution of conflicts linked to gentrification, economic vulnerability, and shifting governance, while contributing a replicable, scalable methodology for studying complex urban phenomena at the intersection of AI, spatial analysis, and social science.

Introduction

Urban conflicts are a fundamental component of city dynamics, emerging from the divergent interests of social, economic, and political actors involved in the production and governance of urban space (Harvey, 2012). Arising from the interactions that structure both urban ecosystems and social life (Janelle and Goodchild, 1984), urban conflicts surface when incompatible interests collide within shared urban spaces (Pelletier and Trudelle, 2015). Such events lead to complex socio-spatial interactions and evolve over time, making their analysis particularly challenging (Razvan, 2020). These conflicts are not only social and political; they are embedded in the very structure of the urban environment and the ecological relations it nurtures. This aligns with the hypothesis posited by geographers since the mid-1970s that cities are the result of ongoing conflicts, with their spatial structure continually shaped by these tensions (Cox and Johnston, 1982; Janelle and Millward, 1976). However, the ecological framework traditionally used to explain urban conflict distribution does not fully capture the complexity of this process. The relationship between urban structure and the conflicts occurring within it is bidirectional and increasingly complex when considering long-term, large-scale conflict networks (Del Romero Renau and Trudelle, 2012; Trudelle and Claramunt, 2024).

Building on the work of urban geographers, recent studies have shown that urban conflicts are deeply intertwined with political, economic, and social variables, which can be modelled through both theoretical frameworks and practical data (Dawwas, 2014; Tiafack and Mbon, 2017). Urban problems, such as housing and environmental inequalities, which refer to the way in which environmental changes, both global and local, differentially affect social groups, particularly in relation to their income levels. According to Castells (2021), urban conflicts emerge as a response to these tensions, with the governance structures of neoliberalism playing a pivotal role in shaping housing policies and urbanization strategies. The relationship between urban conflicts and time has been explored by several scholars, emphasizing how conflicts, when mapped over extended periods, reflect changes in political structures, social movements, and even urban morphology (Cox and Johnston, 1982; Trudelle and Claramunt, 2024). It becomes evident that housing-related conflicts are not static; they develop alongside shifting urban landscapes, creating new spaces of resistance and mobilization.

Recent scholarship highlights how housing conflicts are increasingly driven by the financialization of rental markets in Canada (August and Walks, 2018) and by the shift toward housing privatization and commodification (Soederberg, 2021). These conflicts often arise from the transformation of rental properties into financial assets (Revington and August, 2020). Such dynamics are further shaped by digital technologies mediating urban governance (Hu and Zhang, 2023) and by post-crisis accumulation regimes that reconfigure urban space (Fields, 2022). Building on these insights, our study introduces a computational framework to trace how these macro processes materialize in local housing conflicts, actor networks, and spatial patterns of conflict. While urban housing conflicts have been explored from legal, political, and spatial perspectives, few studies have attempted to model them as multi-layered, evolving networks embedded in both text and geography (Trudelle and Claramunt, 2024). Most existing work remains limited to either qualitative narrative analysis or GIS-based mapping of isolated events. As such, there exists a methodological and theoretical gap in understanding how these conflicts are structured, how they evolve, and how their dynamics may be analyzed at scale. To address this gap, the study is guided by the following question: How can urban housing conflicts be systematically extracted, classified, and analyzed across time and space using AI-based methods, and what do their structural patterns reveal about the socio-political dynamics of housing struggles in Montréal? Our approach aims to bridge gaps in urban related housing studies by providing an integrated representation of urban housing conflicts through the combined use of LLMs, graph-based models, and GIS methodologies. By systematically extracting, validating, and structuring socio-spatial data, we develop an analytical framework that enhances the understanding of housing crises and urban tensions. We examine the role of key actors, government officials, community organizations, and real estate developers, to understand the divergences in how these conflicts are generated. By understanding how urban housing conflicts unfold and evolve, this research seeks to uncover the broader socio-political dynamics at play in shaping Montreal’s housing landscape. By answering this question, we aim to develop and validate a reproducible framework that integrates AI, graph theory, and spatial analysis for the study of urban conflicts. The framework adopted underscores the importance of examining the relationships between public authorities and collective action in the field of urban housing.

The next section introduces the methodological background and principles, along with the iterative LLM-based data-extraction approach. The following Section reports the results of the analyses, including the spatial and temporal classification of conflicts and the identification of the actors involved. Then we discuss the methodological and theoretical contribution and the main findings. Finally, we conclude the paper and outline avenues for future research. 

Methodological framework

Conceptual and methodological foundations

Graph databases offer a compelling alternative (Aldwairi and Jarrah, 2022), organizing entities as nodes and their connections as edges, which enables the modeling of sequential, spatial, and relational dimensions. These methods, widely used in social network analysis (Camacho et al., 2020) and applied in urban contexts such as pedestrian movement (Karim et al., 2021), urban routing (Hristoski and Malenkovsa-Todorova, 2021), and socio-professional mapping (Rahimi et al., 2021), provide valuable alternatives for modelling and understanding complex networks. However, graph databases alone cannot fully address the spatial and contextual complexity of urban conflicts. Their integration with GISs bridges this gap, combining spatial analysis with network modeling to support more nuanced insights. In parallel, the rapid development of LLMs like those by OpenAI (2023) and Brown et al. (2020) enables the extraction of structured socio-spatial knowledge from vast unstructured datasets. LLMs can identify entities, relationships, and spatial-temporal markers in textual sources, enriching both graph and GIS representations with contextual data (Aldwairi and Jarrah, 2022; Bender et al., 2021; Bommasani et al., 2021; Raffel et al., 2020).

Data extraction and structuring process

This LLM-driven approach dynamically generates and adjusts queries to address the complexity and variability of unstructured urban data. The sampling strategy was designed to ensure diversity, breadth, and relevance, incorporating mainstream media outlets, community-based publications, municipal archives, and academic repositories to capture diverse perspectives on housing conflicts in Montréal (cf. complete list in the supplemental material). Future refinements will incorporate more systematic sampling techniques, such as stratification by media type, temporal coverage, and geographic focus. Once extracted, the information was formatted into a structured dataset and validated through cross-referencing with external sources to ensure reliability and reduce bias. A multi-step preprocessing pipeline was implemented to guarantee data quality and consistency. Near-duplicate entries, often resulting from repeated coverage of the same conflict across sources, were grouped and merged based on shared temporal markers, geographic references, and actor involvement. Conflicting information, such as discrepancies in event dates or actor roles, was manually reconciled through cross-referencing. Each conflict instance was enriched by linking associated actors and spatial locations, and the resulting dataset was exported in three formats, CSV, JSON, and Neo4j script, each serving distinct analytical purposes. The structured dataset contains, for each conflict instance, a unique identifier, conflict type, short description, borough name, location coordinates, time period, actors involved, and URL source references. CSV files store structured tabular data for spreadsheet or relational database analysis, JSON files capture hierarchical relationships for integration into Neo4j, and Neo4j scripts enable graph-based modeling of conflicts, actors, and locations.

Two LLM platforms were employed: Copilot for rapid generation and testing of prompt variants, and ChatGPT for context-aware extraction and semantic structuring. This combination allowed nuanced extraction across heterogeneous sources and improved the accuracy of entity recognition and relationship mapping. The analytic framework follows a clear sequence of five steps: (1) compiling textual datasets from diverse sources; (2) developing an iterative LLM-based prompting strategy to identify neighborhoods, (3) classify housing conflicts into major types, and extract detailed attributes such as location, time period, category, and actors involved; (4) processing unstructured text to identify conflicts, actor networks, and spatial-temporal references; and (5) validating extracted data through cross-referencing with municipal reports, advocacy publications, and government datasets.

The extracted data is automatically formatted into three file types: CSV, JSON, and Neo4j script. Each file serves a distinct analytical purpose. CSV files store structured tabular data, processed in spreadsheet applications or relational databases. This includes information on housing conflict events, key actors, actor-conflict links, and geolocation details. JSON files capture hierarchical relationships for integration into Neo4j, a graph-based database supporting network analysis. The resulting Neo4j graph consists of nodes representing conflicts, actors, and locations, with edges denoting relationships such as involvement and spatial or thematic links.

Graph-based and spatio-temporal analysis

The city of Montreal is made up of 19 boroughs, each with its own social, demographic, economic, and geographic characteristics. A taxonomy of Montreal’s urban conflicts was automatically generated structuring the data into 11 types of conflicts (i.e., gentrification and evictions, industrial conversions into luxury housing, controversial real estate projects, short-term rentals, homelessness and social housing deficit, renovictions, real estate speculation, health and safety issues, rent increase, housing shortage, and healthiness). The 11 conflict types were derived through an iterative LLM-based clustering process that extracted candidate categories from the corpus using LLM-driven entity and topic detection, grouping them based on semantic proximity, and then refining these clusters through multiple validation rounds to ensure conceptual coherence. This approach allowed us to capture both dominant themes and nuanced distinctions in housing-related conflicts, avoiding overly broad categories that could obscure important variations. The selection of 11 conflict types reflects a modeling approach designed to preserve analytical granularity and capture the diversity of discourses among involved actors. Aggregating these categories would risk obscuring the specific motivations behind housing-related contention. Similarly, the decision to use 2020 as a temporal boundary is grounded in both empirical and theoretical considerations. Quebec’s housing crisis, which began in 2001 due to supply–demand imbalances following the federal withdrawal from social housing and demographic shifts, re-emerged around 2020 (Careless, 2020). This resurgence was driven by deeper structural issues in the private housing market, compounded by the 2008 financial crisis, the COVID-19 pandemic, and intensified residential development, all contributing to rising rents and renewed conflict.

The crisis today is broader and more complex, marked by eviction moratoriums, emergency shelter policies, new actors like public health agencies and mutual aid groups, and a visible rise in homelessness, notably through growing encampments in Montréal. Theoretically, this turning point aligns with the concept of crisis urbanism, where systemic shocks reshape governance and deepen socio-spatial inequalities. These dynamics are reflected in the conflict network, which becomes more interconnected and city-wide after 2020, making it a meaningful boundary for comparative analysis. The 2001–2024 timeframe was selected to capture key shifts in housing policy and governance in Montréal. This period reflects the influence of neoliberal reforms and includes major milestones such as the continuation and eventual termination of the AccèsLogis program, the adoption of the “Règlement pour une métropole mixte” in 2021, and the launch of the Programme d’habitation Québec in 2022. These changes significantly shaped the local housing landscape and justify the chosen temporal scope (Ducharme et Vaillancourt, 2012; Rose et al., 2022).

Figure 1 shows the general structure of the conflicts identified in relation to the 11 types and actors involved. The layout was automatically arranged by Neo4j′s visualization engine, which positions nodes based on the strength and number of their connections. This results in a natural clustering, where closely related elements appear nearer to each other, making it easier to identify central actors and dominant conflict types. The graph structure reveals distinct patterns that help characterize the dynamics of housing conflicts in Montreal. The clustering of conflicts related to gentrification and evictions highlights systemic tensions within urban space. These hubs show that certain types of conflicts are both more frequent and more interconnected, involving recurring actors and specific locations. The patterns observed indicate that these conflict types tend to co-occur and concentrate spatially, suggesting that they can be effectively modeled at an appropriate level of granularity to capture their relational and territorial dynamics. The presence of multiple types also allows for an analysis of overlap, where some conflicts might fall into multiple thematic areas, highlighting the complexity of housing issues. Similarly, conflicts categorized under Tenant Rights Violations might connect to cases where landlords have used “renoviction” as a strategy to remove tenants and increase rents. The presence of multiple types also allows for an analysis of overlap—some disputes may fall into multiple thematic areas, such as conflicts involving both Social Housing Shortages and Municipal Policy Changes, indicating deeper structural issues.

Figure 1.

Types, conflicts, and actors.

The conflict association graph reveals structural patterns in how different housing disputes are interconnected through shared actors (Figure 2). Using Neo4j, each conflict is represented as a node, and links indicate common elements. The layout automatically groups closely related conflicts, helping to reveal clusters and recurring patterns across the dataset. Conflicts involving major advocacy groups like City of Montreal tend to cluster together (left cluster), as these organizations repeatedly engage in cases related to evictions, rent control, and social housing shortages. The recurring presence of provincial advocacy groups within conflict clusters illustrates the persistence of urban social movements, where marginalized actors mobilize against structural inequalities in housing.

Figure 2.

Conflicts relationships.

Figure 3 illustrates the main conflict communities, identified as clusters, highlights key structural patterns in Montreal’s housing disputes. Each cluster represents a group of interconnected conflicts, potentially sharing common actors, themes, or geographic locations. To detect these communities, we applied the Louvain algorithm, a method specifically designed for optimizing modularity in large networks (Blondel et al., 2008). This algorithm effectively identifies densely connected groups of actors, revealing patterns of association that might otherwise remain obscured within the broader conflict landscape. For example, one prominent cluster centers on tenant rights and evictions, where conflicts involving tenant advocacy organizations and provincial advocacy groups frequently emerge, indicating a strong network of resistance against displacement. Another cluster is tied to real estate development and gentrification, with repeated involvement from major developers, often linked to urban renewal projects and municipal zoning decisions (Annunziata and Rivas-Alonso, 2022). Additional clusters reflect neighborhood-specific struggles, such as housing affordability debates or redevelopment tensions. The density and connectivity within each cluster reveal how tightly conflicts are interwoven, while bridging nodes highlight disputes that span multiple issues—such as municipal policy changes affecting both social housing and urban development. Analyzing the composition of these communities helps expose persistent fault lines in Montreal’s housing landscape and the key actors shaping its ongoing struggles over space and affordability.

Figure 3.

Conflicts, actors, and communities.

Temporal patterns reveal how housing conflicts evolve over time. The 11 types of conflicts identified by the use of LLM are represented, including gentrification processes, evictions, real estate speculation, rising rent prices, the shortage of social housing, and the housing crisis It uses timeline attributes stored in the graph database to display conflict evolution across two periods, before and after 2020. A temporal filtering query was applied to separate clusters by time. Some conflicts are tied to specific policy shifts, such as the introduction of new short-term rental regulations in response to the impact of Airbnb on housing availability, which sparked tensions in central neighborhoods. Other disputes are part of long-term structural trends, such as the gradual reduction of affordable housing in areas near universities and in multiethnic neighborhoods, where international students and immigrant communities face increasing difficulties in securing stable housing. Looking at past cases, we can also trace how major urban renewal projects, such as the demolition of low-income housing units in the 1990s to make way for luxury condominiums, continue to shape current housing struggles. Time-series analysis could further reveal whether specific periods such as those following government budget cuts to social housing programs, correspond with spikes in conflict activity. This would allow us to identify historical flashpoints and potentially anticipate future areas of contention based on emerging trends.

The network of actors also shifts, reflecting changes in alliances and power dynamics. The shift from modular to interconnected conflict networks post-2020 reflects a transition from localized disputes to systemic urban crises, aligning with ecological theories that view cities as dynamic systems shaped by long-term socio-political transformations. Developers and municipal policy makers remain influential, but new coalitions form, suggesting a broadening of housing activism beyond traditional advocacy groups. The nature of conflicts evolves as well. Before 2020, disputes were primarily tied to gentrification, rent control, and social housing policy, focusing on neighborhood-level urban renewal projects and long-term affordability concerns. Post-2020, conflicts increasingly involve pandemic-related emergency housing measures, mass evictions due to job losses, and legal battles over rent moratoriums. The graph structure suggests that conflicts after 2020 are more interconnected, as struggles over eviction protection, homelessness policies, and affordability regulations frequently overlap, creating a denser network of disputes.

Structurally, the conflict graph also changes over time (Figure 4). Before 2020, it is more modular, with tightly knit clusters representing localized disputes in specific boroughs. After 2020, the structure becomes more interconnected, with conflicts linking across different neighborhoods and policy domains, likely due to shared struggles emerging from a system-wide crisis. This shift suggests that housing conflicts are no longer just shaped by neighborhood-specific tensions but increasingly driven by broader economic instability and policy responses. The key transformation after 2020 is the expansion and intensification of housing conflicts, reflecting the growing systemic nature of housing struggles in Montréal. These findings indicate that future conflicts may continue to be shaped by macroeconomic crises and policy shifts rather than localized redevelopment disputes alone.

Figure 4.

Conflicts temporal distribution.

The comparative analysis of housing conflicts before and after 2020 reveals significant shifts in patterns, actors, and thematic focus, reflecting broader socioeconomic and policy changes (Figure 5). A temporal filtering query was applied to separate clusters by time and borough-level layers were combined with categorized conflicts to show changes in spatial concentration over time. Prior to 2020, conflicts were relatively localized and fragmented, concentrated in long-standing gentrification hotspots. These disputes primarily involved tenant advocacy groups resisting rising rents and evictions linked to gentrification. After 2020, however, the frequency of conflicts increased, likely due to the economic strain caused by the COVID-19 pandemic, which intensified housing precarity, mass evictions, and affordability crises across a broader range of boroughs. As a result, housing-related disputes expanded beyond traditional high-pressure areas, affecting a wider spectrum of communities. These conflicts include controversial real estate developments, gentrification and eviction processes, issues related to unsafe and substandard housing, the conversion of industrial brownfields into luxury residences, real estate speculation, and renovictions. These conflicts are primarily concentrated in the central and southeastern boroughs of Montreal. This concentration can be explained by the dynamism of the real estate market in these areas, coupled with urban revitalization policies implemented since the early 2000s.

Figure 5.

Conflicts spatial distribution before 2020 and after 2020.

Figure 5 also reveals a strong concentration of conflicts in central and eastern boroughs, which can be attributed to both heightened real estate pressure and active civic engagement in these areas. Two key types of conflict stand out. First, issues related to homelessness and the shortage of social housing are particularly prevalent. These reflect persistent barriers to accessing affordable housing and an insufficient supply of social housing in densely populated boroughs. This situation has been further exacerbated by the COVID-19 pandemic, which deepened the vulnerability of at-risk populations, despite municipal efforts to address housing challenges. Second, conflicts linked to short-term rentals, especially those involving platforms like Airbnb, are widespread. These signal a shift in the rental market, where long-term housing is increasingly converted into tourist accommodations, thereby reducing availability for local residents.

Discussion

Methodological insights

The combined use of LLMs, graph databases, and GIS provides a scalable and theoretically grounded approach to urban conflict analysis, overcoming the limitations of traditional methods that rely on isolated case studies or simple spatial mapping. LLMs extract rich contextual information from unstructured sources, graph databases model evolving relationships among actors, events, and places, and GIS situates these dynamics in space and time. This integration enables the identification of structural patterns, actor coalitions, and spatial hotspots, offering a more nuanced understanding of power relations in housing conflicts.

A key contribution of this framework is its ability to represent diverse actors—from real estate developers to municipal agencies and advocacy groups, while highlighting the spatial areas most affected by these tensions. Furthermore, the study demonstrates how LLMs can be incorporated into a structured knowledge extraction process without extensive domain-specific pre-training. Unlike conventional NLP pipelines, our iterative prompting strategy leverages the generalist capabilities of LLMs while refining outputs to capture complex spatial, temporal, and relational dimensions of urban conflicts.

Conflict and control in Montreal’s housing landscape

The clustering of conflicts around gentrification and evictions reflects Castells’ conception of urban space as a battleground for competing social interests, where neoliberal urban policies collide with grassroots resistance (Castells, 1972; Harvey, 2008). These clusters reveal systemic tensions: conflicts are not only frequent but highly interconnected, often involving recurring actors and locations. This pattern underscores that housing disputes are embedded in broader struggles over urban governance and territorial control. The spatial concentration of conflicts supports ecological conflict models (Cox, 2024; Janelle and Millward, 1976), which posit that urban morphology is shaped by socio-economic pressures and contested land use.

Capitalist initiatives have driven densification, the conversion of industrial lands, and the emergence of large-scale real estate projects, often financed by investment firms seeking high returns. These boroughs have become prime targets for residential development, resulting in rising rents and intensifying land pressure. Between 2000 and 2015, real estate developers played a dominant role in housing production, yet this capital-driven development remained largely inaccessible to low- and moderate-income households, exacerbating their exclusion from the private rental market (Gaudreau et al., 2021). Eight boroughs were selected for integrated urban revitalization programs, which significantly reshaped their built environment. In areas such as Sud-Ouest and Villeray–Saint-Michel–Parc-Extension, these interventions accelerated real-estate pressure and triggered sharp increases in housing prices.

Another factor influencing the spatial distribution of conflicts is the presence of active advocacy networks. These organizations have been instrumental in exposing abuses and mobilizing resistance against housing injustices in Montreal. They organize protests, challenge rent hikes, and advocate for better living conditions, forming a strong network of resistance to displacement. Municipal authorities and provincial agencies also hold significant influence through zoning laws, housing policies, and regulatory measures. However, they are frequently criticized for slow responses or insufficient action. Political actors such as Projet Montréal—the municipal party in power since 2017—prioritize affordable housing and homelessness, working to regulate urban development and protect tenants. Yet they face major challenges, particularly due to limited support from the provincial government. This interplay of actors illustrates the complexity of governance in Montreal’s housing landscape.

Our analysis also reveals neighborhood-specific struggles across the city, though they are particularly acute in certain areas—for example, affordability debates in Hochelaga-Maisonneuve and redevelopment-related tensions in Le Plateau–Mont-Royal. The density and connectivity within each conflict cluster demonstrate how deeply interwoven these disputes are, while bridging nodes between clusters highlight issues that span multiple domains, such as municipal policy changes affecting both social housing and urban development. Analyzing the composition of these communities reveals persistent fault lines in Montreal’s housing system and the key actors shaping the city’s ongoing struggles over space and affordability. These processes substantially increase land values, which in turn drive up housing costs. Time-series analysis could further reveal whether specific periods saw spikes in conflict—such as following government budget cuts to social housing programs—offering a basis for predicting future events based on historical trends.

The comparative analysis of housing conflicts before and after 2020 reveals significant shifts in patterns, actors, and thematic focus, reflecting broader socioeconomic and policy changes. Prior to 2020, conflicts were relatively localized and fragmented, concentrated in long-standing gentrification hotspots within central neighborhoods. These disputes primarily involved tenant advocacy groups resisting rising rents and evictions driven by gentrification. After 2020, however, the frequency of conflicts increased, likely due to the economic strain caused by the COVID-19 pandemic, which intensified housing precarity, mass evictions, and affordability crises across a wider range of boroughs. As a result, disputes spread beyond traditional high-pressure areas, affecting more diverse communities.

This evolution reflects broader neoliberal dynamics shaping major cities today, characterized by the financialization of housing and a retreat of public intervention in the sector. In particular, the federal government’s withdrawal from social housing programs since the 1990s has created a structural imbalance in supply, deepening socio-spatial inequalities and contributing to rising homelessness in Montreal. Rising land values and rental costs have made both homeownership and renting increasingly inaccessible, not only for low-income households but also for segments of the middle class, pushing vulnerable populations out of central urban areas and away from essential services. These findings resonate with wider debates in urban studies concerning the financialization of rental housing markets. Moreover, the use of LLMs to extract and classify policy narratives aligns with emerging work on how digital technologies shape urban governance and public discourse. Finally, the shift toward more interconnected, city-wide conflict networks after 2020 supports theories of post-crisis urban restructuring, where systemic shocks catalyze new forms of spatial inequality and political contestation.

Beyond its methodological contributions, this study advances theory by operationalizing concepts from urban social movement theory and territorial struggle through the joint analysis of temporal and spatial conflict patterns. Our multidimensional approach makes it possible to analyze complex urban phenomena systematically. While regime theory and growth machine analysis highlight elite-driven urban development, our graph-based and spatial methods reveal how these dynamics are reflected in the evolving structure of housing conflicts. By tracing actor networks, spatial clusters, and temporal trends, the study grounds abstract theories in large-scale empirical evidence and offer a replicable framework for analyzing urban governance and resistance. By analyzing temporal shifts in conflict patterns, the study also uncovers how actors’ influence has evolved, especially in response to major socio-political events such as the 2006 economic crisis and the emergence of short-term rental platforms like Airbnb. In the context of Montreal, this approach proved particularly effective for navigating heterogeneous, large-scale open archives, media reports, and advocacy documents without relying on rigid lexicons.

Pending challenges

However, the method also faces further challenges. While this study does not include formal inter-rater reliability metrics or systematic benchmarking against external datasets, several validation steps were implemented to ensure data robustness. These include iterative prompt refinement, manual verification of selected outputs, and cross-referencing with multiple open data sources. Future work will incorporate inter-annotator agreement measures, structured comparisons with curated datasets, and methods for reconciling conflicting narratives across sources. The reliance on public textual data makes the extraction process sensitive to narrative biases embedded in news outlets, political discourses, and advocacy reports. While validation steps such as manual verification and cross-referencing with multiple sources were implemented, a series of biases can influence which events are reported, how actors are portrayed, and what narratives are emphasized. Although our iterative prompting strategy helps surface diverse perspectives, future methodological enhancements should include systematic bias detection techniques such as source classification and triangulation across ideologically diverse outlets—to better account for these influences in conflict identification.

Although LLMs demonstrate strong contextual reasoning, they struggle with conflicting narratives and lack inherent source attribution mechanisms (Adolphs et al., 2022; Zhang et al., 2022), necessitating systematic cross-validation and manual triangulation to ensure analytical robustness. Future refinements in query design should aim to differentiate how advocacy groups, government agencies, and real estate developers describe conflicts, recognizing potential biases embedded in these narratives. The study of major historical events, such as the 2006 financial crisis in the context of this study, will be crucial for understanding how economic shocks influence the spatial and temporal evolution of housing conflicts, revealing shifts in policy responses and market dynamics.

Further research will also explore the evolution of conflict patterns in both space and time, examining how their magnitude fluctuates across different neighborhoods and historical periods. While the current methodology ensures spatial-temporal attribution, further refinements should incorporate techniques to quantify the magnitude of conflicts over time. This can involve trend detection algorithms that analyze frequency and intensity, highlighting emerging patterns and persistent issues across different neighborhoods. This includes analyzing the ways in which proximity, both physical and temporal, affects public discourse and media reporting, investigating whether individuals and institutions closer to the conflict sites describe them differently compared to distant observers. An extension of this methodology should investigate how narratives shift over time and space, analyzing whether discourse changes when conflicts are reported from a local, national, or historical perspective.

Conclusion

This paper introduced a novel methodological framework for analyzing urban housing conflicts in Montreal by integrating LLMs, graph databases, and GIS. Through the extraction and structuration of conflict data from unstructured sources, the study mapped the evolution of housing disputes across time and space, identifying key actors, conflict types, and spatial hotspots. The findings reveal how housing tensions are shaped by gentrification, speculative investment, and governance dynamics, offering a replicable approach for studying complex urban phenomena. This contribution enhances the empirical understanding of housing struggles and provides a scalable tool for urban analysis.

Theoretically, the study bridges computational methods with critical urban theory, particularly Castells’ theory of urban social movements and ecological models of urban structure. This pragmatic approach calls for attention to the social contexts in which researchers produce knowledge by combining a variety of fields. By modeling conflicts as dynamic, multi-layered networks embedded in discourse and geography, the research advances the conceptualization of urban conflicts as systemic and evolving. It demonstrates how AI-driven tools can operationalize theoretical constructs, enabling the analysis of socio-political tensions at scale. This synthesis of theory and computation contributes to a deeper understanding of how urban inequalities are produced, contested, and spatially manifested. Rather than replacing these theories, our methodology provides a means to empirically test and refine them using large-scale, multi-source data. It also highlights the evolving nature of urban conflict in the context of systemic crises, such as the COVID-19 pandemic, where traditional governance models are disrupted and new coalitions emerge. This dynamic perspective contributes to a more nuanced understanding of how urban inequalities are produced, contested, and reconfigured over time. The approach presented is not city-specific. Its modular design allows researchers and planners to replicate the workflow in other urban settings, using locally available data sources. Future applications could include comparative studies across cities or thematic extensions to other domains of urban governance and conflict.

Future research should expand this framework to other cities and conflict types, exploring its adaptability across different socio-political contexts. Enhancing the interoperability between LLMs, graph databases, and GIS platforms could support real-time monitoring and participatory planning. Further work should also investigate the narrative framing of conflicts by different actors, the role of historical events in shaping conflict trajectories, and the integration of socio-economic indicators to assess vulnerability. By refining and extending this approach, researchers and policymakers can better anticipate, understand, and respond to the evolving landscape of urban housing conflicts.

Supplemental material

Supplemental Material—Extracting and analyzing urban housing conflicts using large Language models, graph databases and GIS

Supplemental Material for Extracting and analyzing urban housing conflicts using large Language models, graph databases and GIS by Monique Mato, Catherine Trudelle, Christophe Claramunt, and Eliott Libner in Environment and Planning B: Urban Analytics and City Science