Urban data: harnessing subjective sociocultural data from local newspapers

SPECIAL COLLECTION: DATA POLITICS IN THE BUILT ENVIRONMENT RESEARCH CORRESPONDING AUTHOR: Filipe Mello Rose Digital City Science, HafenCity University Hamburg, Hamburg, DE; Laboratory of Knowledge Architecture, Technical University Dresden, Dresden, DE [email protected] KEYWORDS: cities; data-based governance; data politics; datafication; local journalism; planning; sociocultural data; subjective data; urban data; urban governance TO CITE THIS ARTICLE: Mello Rose, F., & Chang, J. (2023). Urban data: harnessing subjective sociocultural data from local newspapers. Buildings and Cities, 4(1), pp. 369–385. DOI: https://doi.org/10.5334/bc.300 Urban data: harnessing subjective sociocultural data from local newspapers FILIPE MELLO ROSE JUIWEN CHANG ABSTRACT As data-based governance becomes mainstream, social and cultural interactions that characterise urban life are at risk of being ignored in decision-making practices if only supposedly objective, quantifiable data are used. In this context, this article conceptualises subjective sociocultural data as a data form that considers a city’s intangible and unquantifiable social and cultural aspects. A methodology is proposed for collecting and using subjective sociocultural data by highlighting local press articles as a potential data source. A pilot application conducted in Hamburg, Germany, demonstrates a potential integration of subjective sociocultural data into urban planning processes by analysing over 2500 local newspaper articles. The findings reveal that local journalism can be a data source for understanding diverse social and cultural interactions between citizens and urban places. This street-level information from local newspaper articles can (1) provide urban planners with an overview of newspaper mentions of any specific urban areas, (2) support the identification of local debates, and (3) aid in the observation of emerging places of sociocultural interactions. This approach can support the diverse government and non-government stakeholders engaged in data-based governance to better account for intangible sociocultural aspects of urban life. PRACTICE RELEVANCE This research supports governance actors in dealing with the epistemological limitations of purposely gathered and/or objective data by conceptualising a new—currently untapped—data type: subjective sociocultural data sourced from local journalism. By using geographical text analysis on local newspaper articles, urban planners and decision-makers gain access to a wealth of street-level information, local debates and temporal dynamics of urban issues. This approach provides a comprehensive understanding of intangible and unquantifiable aspects of urban life, allowing for more informed and context-sensitive decision-making. The practical benefits include identifying diverse uses of urban spaces, capturing local public debates, and tracking the emergence and disappearance of places in the public sphere, possibly leading to more effective and inclusive urban planning practices. *Author affiliations can be found in the back matter of this article 370Mello Rose and Chang Buildings and Cities DOI: 10.5334/bc.300 1. INTRODUCTION: BETTER DATA FOR URBAN GOVERNANCE Many new trends in urban planning, such as algorithmic governance, smart cities and digital twins, rest on data-based urban governance. In this type of urban governance, ‘a complex system of actors, relationships, processes, and technologies’ is designed to decide on urban issues based on diverse sets of available and analysable data (Maffei et al. 2020: 124). Data-based planning is a crucial instrument for addressing the upcoming environmental and societal challenges (Batty & Yang 2022: 146) as it allows comprehensive and interdisciplinary analyses of complex problems by making them ‘more “legible” for governing’ (Mejias & Couldry 2019: 4). At the same time, the datafication of urban governance raises various issues concerning data validity, as well as the reliability, representativeness and interoperability of available data (Bunders & Varró 2019). Crucially, data-based urban governance is limited by all urban problems not being ‘equally knowable and solvable’ by data-based governance practices (Bunders & Varró 2019). Many key intangible and unquantifiable aspects of urban life (e.g. the role of family-owned corner shops in a local neighbourhood) play a vital role in the day-to-day urban experiences of citizens (Jacobs 1961/1992). Likewise, urban places such as parks or pavements ‘mean nothing divorced from their practical, tangible uses’ (Jacobs 1961/1992: 111). This information is hardly quantifiable and is thus ignored in data-based governance. These limitations of data-based governance can at least be mitigated by including broader and more variegated datasets (Batty 2019). In this sense, the present article conceptualises subjective sociocultural data as an important yet frequently ignored or underused input for data-based governance. With this conceptualisation and an exemplar pilot methodology, the aim of this paper is to support the diverse stakeholders engaged in data-based governance (including government and non-government actors) to better account for intangible and unquantifiable aspects of urban life in their decision-making. Subjective data are a particular data form that explicitly involves human judgment in its collection. Sociocultural data depict aspects ‘related to the different groups of people in society and their habits, traditions, and beliefs’ (Cambridge University Press n.d.). Jointly, subjective sociocultural urban data depict self-reported and evaluated accounts of a city’s social and cultural practices (i.e. the interactions forming a city’s social and cultural fabric). In pursuing the research goal of broadening the data sources for urban governance, this article conceptualises the utility of subjective sociocultural urban data and proposes a methodology for collecting and using this data type. This conceptualisation and methodology rest on a critical review of the literature on the different types of data used in urban governance and a synthesis of conceptual discussions on subjective data. Moreover, local press articles were identified as a potential and largely untapped source of subjective sociocultural urban data, which, once geo-referenced and synthesised, can inform local governance decisions. Empirically, this study reflects on a pilot study that aims to integrate subjective sociocultural data into urban planning by drawing on a research project carried out with the state-owned company for real estate management and land assets of the German city of Hamburg: the Landesbetrieb Immobilienmanagement und Grundvermögen Hamburg (LIG). The pilot study developed a geoparsing module with subjective sociocultural data that aims to complement ‘objective’ data used in decision-making regarding public real estate policies. This study draws on data from 2539 local newspaper articles from Hamburg published online between 2018 and 2022. The articles were automatically analysed for spatial references in Hamburg (i.e. geo-parsed and geocoded) and clustered around 10 algorithmically generated topics. The pilot use cases show that local journalism can (1) provide urban planners with an overview of newspaper mentions of any specific urban areas, (2) support the identification of local debates, and (3) aid in the observation of emerging places of sociocultural interactions. In conceptualising local press articles as a source of subjective sociocultural urban data for urban governance, this study draws on and contributes to research on datafication processes in governance. However, as ‘datafication is cross-disciplinary in nature’ (Lomborg et al. 2020), this study also draws on multiple academic fields. First, it draws on media studies to argue for the general utility of (local) journalism as a data source for the data-based governance (Battistini et 371Mello Rose and Chang Buildings and Cities DOI: 10.5334/bc.300 al. 2013; Sarın & Uluğtekin 2019). Second, it draws on digital natural language processing (NLP) methods for geographical text analysis and topic modelling (Cai 2021; Middleton et al. 2018; Porter et al. 2015). This article is structured as follows. Following this introduction, the paper conceptualises and situates subjective sociocultural data among other urban data forms used for urban planning. The methodology is then described and the results of an empirical pilot application are reported. This empirical section sets out the NLP methods and three potential use cases of the outlined techniques. The last section reviews and discusses the use of subjective sociocultural data for urban governance. 2. EXPLICITLY SUBJECTIVE DATA FOR DATA-BASED GOVERNANCE 2.1 FROM GOVERNMENT DATA TO DATA-BASED GOVERNANCE In contrast to the traditional governments’ use of self-collected data, data-based governance draws on a variety of data sources. These range from outcomes of algorithms and the Internet of Things (Coletta & Kitchin 2017), as well as large datasets, dashboards or surveillance systems to create a more efficient administration and governance of places (Bunders & Varró 2019; Kitchin & McArdle 2017; Valdez et al. 2018; Zuboff 2019). Moreover, with the increasing prevalence of sources in data-based governance, many stakeholders are able and needed to provide or analyse data. Most notably, international platform corporations now also inform decision-making and policy development with their datasets and analytical tools (e.g. Rettberg 2020). Along with numerous other factors, this datafication has thus further advanced the shift from traditional government to governance by beyond-the-state ‘horizontal associational networks of private (market), civil society […] and state actors’ (Swyngedouw 2005: 1992). Data-based governance, thus, coordinates this ‘beyond-the-state governing’ arrangement by providing and analysing significant and diverse datasets. This datafication improves the capacity for anticipatory planning in a governance system that includes an ever greater variety of actors (beyond traditional government administrations) by using data ‘to revolutionize the process of policy analysis’ (Maffei et al. 2020: 124). While databased governance can thus represent new business opportunities for companies collecting data as a by-product (e.g. Rettberg 2020), it can also allow researchers or civil society organisations to put governments and corporations under greater scrutiny (e.g. Dalton 2019). Crucially: datafication enables us to deal with lines of inquiry that were difficult if not impossible to pursue before. (Lomborg et al. 2020: 207) In this sense, data-based governance can also foster a more collaborative and inclusive approach to governance, where a network of actors work together to identify and address complex societal challenges. By providing and analysing data that are of interest to urban governance, these nongovernment actors thus, directly or indirectly, engage in expanding networks of data-based governance. 2.2 ONTOLOGY OF DATA SOURCES OF DATA-BASED GOVERNANCE Before digitalising public administration, economic and social activities, data-based governance relied primarily on purposely collected data, such as structured surveys and observation data. This type of data source includes data deliberately produced by government institutions (e.g. data from census and statistics offices) or non-government organisations supporting data-based governance with their data. For instance, data-based governance often draws on specialised survey institutes or universities to collect data with diverse formal methods (online surveys, in-person interviews, online tracking, etc.). A trait of these purposely collected data or survey-type data is, however, that its controlled collection processes reduce the data’s ‘scope, temporality, and size, and are quite inflexible in their administration and generation’ (Kitchin & McArdle 2016: 2). 372Mello Rose and Chang Buildings and Cities DOI: 10.5334/bc.300 Digitalising social, economic and governmental activities has enabled the collection of usable data as a by-product. As public administrations become more digitised, more data are processed digitally in government bureaucracies and become a potential new data source (Kitchin & McArdle 2016). This type of data, which are gathered as a by-product, is either observed (i.e. as a result of people using technology) or inferred (i.e. consolidated information from existing data sources) (Taylor & Richter 2015). Data that originate as a by-product of government activities also include records, such as budgets and spending data, and data on assets as well as from registries (e.g. registration of residences, financial activities or vehicles). Crucially, this data source type is not limited to any ex-ante specification of purpose in data-based governance, as the possible knowledge gains are defined after the data are collected during data preparation (Kitchin & McArdle 2016) (Table 1). All data (sources) have natural epistemological limits. Purposely collected data are limited to addressing societal problems known before the data collection based on scientific theory (Kitchin 2014). While mostly free of sampling and statistical errors, this data type has a low potential to fulfil unanticipated usages (Kitchin & McArdle 2016). Unstructured data collected as a by-product, in contrast, are prone to sampling and statistical errors and biases originating from unequal uses of a service or product but allow exploratory research. Traditional data sources in urban governance generally relied on the accessibility of purposely collected data, such as surveys and censuses. As purposely collected data require an ex-ante definition of research problems (i.e. before data collection), data-based governance must also consider data sources allowing greater exploratory research to identify new research problems. 2.3 SUBJECTIVE AND SOCIOCULTURAL DATA Sociocultural data depict ‘society and their habits, traditions, and beliefs’ (Cambridge University Press n.d.). They feature structured and statistical cultural, ethnic, religious or demographic information, as well as (more subjective) data on cultural practices, day-to-day activities and routines, and beliefs. In this article, sociocultural data are essentially data that allow one to understand better a given area’s social and cultural fabric. Data-based governance primarily draws on supposedly ‘objective’ data as impersonal and neutral ‘raw information’ that are best collected with the least ‘subjective’ human involvement possible (Rieder & Simon 2016). This applies to sociocultural data that are often reduced to variables for ethnic, religious or class identification which can be assessed somewhat objectively. Subjective data, in contrast, significantly and explicitly involve human judgment in its production. Despite a greater risk for biases, this data type is used as purposely collected data to measure social phenomena such as happiness, affection or wellbeing (e.g. Kahneman & Krueger 2006; Macků et al. 2020). Subjective measures are relevant for two main reasons. First, in the absence of (more) objective measures, subjective ones can (at least partially) depict social phenomena on which data are scarce. Second, subjective measurements allow one to: capture changes in both the explicit and the implicit components of the variable being measured and, therefore, […] can be better suited for the study of broadly defined concepts. (Jahedi & Méndez 2014: 3) GOVERNMENT DATA NEW DATA-PRODUCERS IN DATABASED GOVERNANCE Purposely collected data (Structured) Periodically collected data on the polity, e.g. census data, surveys of statistics administrations Data from non-state data collection companies and research institutes, i.e. market research, university research Data as a collected as a by-product (Unstructured) Data collected as a by-product of government administrations, e.g. permit data, data from public companies (i.e. health insurance, unemployment insurance, public transport), budget data Data collected as a by-product of ordinary economic and social activities, e.g. social media companies, targeted advertisements, customer reviews Table 1: Conceptualisation of different data sources and collection forms. 373Mello Rose and Chang Buildings and Cities DOI: 10.5334/bc.300 Data on the sociocultural aspects of urban life meet both reasons for using subjective measures. For one, sociocultural factors of urban life offer few alternative ‘objective’ measures. Most urban interactions and practical, tangible uses of places are difficult to quantify objectively without using (sometimes far-fetched) proxy measures. For another, the sociocultural aspects of urban life are a broadly defined concept that refers to the intangible aspects, such as the diverse social and cultural interactions enriching urban life. Allowing different actors to determine the relevant social and cultural interactions will likely better depict the local urban social and cultural fabric. Moreover, while statistical and systematically quantified sociocultural data allow large-scale analyses, smallscale subjective data can help planners to better understand the social conditions shaping an area’s urban (sociocultural) fabric. Urban sociocultural data depict day-to-day activities, routines, values and public spaces’ practical, tangible uses. In this sense, this article understands subjective sociocultural data as collections of descriptions of social interactions and cultural life in an area. This includes accounts of cultural and civic events, the uses of public spaces, or popular meanings of places (e.g. the importance of traditional corner shops or bars to a community) relevant to understanding the local urban fabric. Combining different types of data sources allows for triangulating information and complementing analyses with different perspectives. In this sense, adding data gathered as a by-product to databased governance eases exploratory, unexpected lines of enquiry, while subjective sociocultural data expand data-based governance to better account for a given area’s social and cultural fabric. 2.4 LOCAL JOURNALISM AS A DATA SOURCE FOR URBAN GOVERNANCE This study conceptualises and empirically tests the use of local press articles as a source of subjective sociocultural urban data. In doing so, it draws on multiple studies in which press articles serve as an explicitly subjective source of geographical and sociocultural data. For instance, Voukelatou et al. (2021: 300) find that ‘news data are a new promising data source for the further exploration of subjective well-being’. Battistini et al. (2013: 157) use online news feeds as data sources to efficiently map geohazards (i.e. earthquakes, landslides and floods) under the assumption that whenever geohazards have ‘relevant consequences, news is reported on the Internet’. Gregory & Paterson (2020) elaborate on this practice under the title of geographical text analysis, for which the scholars geo-parse historical press articles to map historical poverty in the UK. The scholars’ study on historical poverty demonstrates how analysing textual data can provide insights into geographical patterns (Gregory & Paterson 2020). Ozgun & Broekel (2021) use regional newspapers to enquire about the regional difference in attitudes to innovation by quantifying the mentions and sentiments of different regional newspaper articles. However, while most studies focus on the national or at least regional level, Paterson (2020: 66) points out that: references to place, especially at more local levels, can make the impact(s) of abstract concepts […] appear more concrete by situating them within definable geographical boundaries. Other studies have used similar methodologies to mobilise social media data for data analysis in environmental (Ghermandi & Sinclair 2019) and tourism research (Chen et al. 2021). However, in contrast to social media data, press articles have at least a minimal degree of quality control. Moreover, while interactions on social media are highly diverse and frequently use colloquial or group-specific language, newspaper articles are typically written for broader audiences and require less context-specific language or knowledge about the authors to be intelligible. In this spirit, applying geographical text analysis to local press articles appears as an auspicious way of introducing new sources of subjective sociocultural data to data-based urban governance. The use of newspaper articles as a source of sociocultural data is grounded on the assumption that the content produced by newspapers at least generally depicts cultural practices, day-to-day activities and routines, and beliefs of a given place. The way media outlets ‘frame an issue shapes how people understand and remember it’ (Ozgun & Broekel 2021: 3). Moreover, media outlets 374Mello Rose and Chang Buildings and Cities DOI: 10.5334/bc.300 have nested interests that influence their framing and categorisations of what is ‘newsworthy’. However, while the press might advance specific agendas, ‘they do not do so independent of their audience’ (3). As most media organisations are commercially driven, the choice of newsworthiness and the tone of reporting news is strongly linked with attempts to gain and retain the largest possible audiences (Scheufele 1999; Gentzkow & Shapiro 2010; Agirdas 2015). As people seem to ‘selectively expose themselves to attitude-consistent information’ (Earle & Hodson 2022: 9), aligning to the political orientation and reporting priorities of a target readership is a commercial necessity and a response to public demand (Gentzkow & Shapiro 2010). While national news outlets tend to become increasingly partisan to win audience loyalty in a politically divided market (Gentzkow & Shapiro 2010), this is not the case for local newspapers due to their limited geographic competition (Agirdas 2015). 3. EMPIRICAL PILOT APPLICATION: SOURCING LOCAL JOURNALISM FOR URBAN GOVERNANCE In the following, this study describes and discusses an empirical pilot application of the possible uses and sources of sociocultural data for data-based governance conceptualised above. While this pilot application has been developed in partnership with a state-owned company for a specific case (i.e. to improve data-based governance of public real-estate assets), the methodology (including the used algorithms) is available as open-source code on GitHub. This way, the practical implementation of sourcing local journalism for urban governance can be tested and improved in other contexts. 3.1 CONTEXT OF THE EMPIRICAL PILOT APPLICATION This study is situated within a broader research project that investigates new modes of data-based planning conducted with Hamburg’s state-owned company for real estate management and land assets, the LIG. The German city of Hamburg is no exception to the trend of datafication of urban governance, and the LIG aims to improve its governance activities with new digital tools for spatial analysis. To develop a method of digital data-based real estate and land management in the city-state, a collaboration between the LIG and HafenCity University has engaged in the creation of a digital platform for the evaluation of land parcels. This platform will consolidate various data sources and inform decisions regarding the acquisition and sale of land by public administrations. More precisely, the digital tool allows the comparison of land parcels based on their surroundings (i.e. the proximity to various urban amenities) and their connectivity (i.e. isochrones in the city) in the ‘LIG-Finder’ module. In addition, the ‘geo-parsing’ module provides a spatial overview of geocoded texts. This geo-parsing module displays geocoded documents from the state parliament and multiple regional and local newspapers. This includes the Elbe Wochenblatt, a hyperlocal free newspaper that this study uses as a pilot application. Elbe Wochenblatt is a hyperlocal advertisement-based newspaper from Hamburg. Its articles are disseminated online and in seven weekly neighbourhood-level print editions, distributed free of charge to all households (unless they opt out) within a specific delivery zone (Figure 1). As an advertisement-based newspaper, it provides a platform for and depends on local businesses to advertise their products and services. Elbe Wochenblatt features short articles using colloquial language on local politics, sports, civic life, cultural events and businesses. Next to Hamburger Wochenblatt, Elbe Wochenblatt is the second largest free newspaper in Hamburg with approximately 300,000 weekly printed copies.1 However, the media landscape in Hamburg is also—and probably more substantially—shaped by regional and Hamburg-based national newspapers, such as MOPO/ BILD, Hamburger Abendblatt and Zeit.2 Elbe Wochenblatt’s accessibility and high volume of publications from 2018 to 2021 make it a valuable test case for the pilot application of the LIG-Finder’s geo-parsing module. During that period, the newspaper was majoritively owned by Madsack Mediengruppe, with a minority stake held by Funke Mediengruppe. In 2021, the newspaper was fully acquired by Funke Mediengruppe, which integrated the previously independent editorial office with other local newspapers from 375Mello Rose and Chang Buildings and Cities DOI: 10.5334/bc.300 Hamburg (notably with Hamburger Wochenblatt) in January 2023. While Madsack Mediengruppe is regionally focused on northern Germany, Funke Mediengruppe is the 10th largest media group in Germany (Institut für Medienund Kommunikationspolitik 2022). This study uses all 3511 newspaper articles published online (and in at least one of the seven neighbourhood print editions) from 2018 to 2021. The time frame was selected due to the newspaper’s above-average publishing activity. The focus is on articles authored by the newspaper’s editorial staff to safeguard a consistent format and focus on the newspaper’s target areas. This excludes submissions and commentaries from readers that vary vastly in format, making an automated corpus analysis difficult at this piloting stage. Elbe Wochenblatt was selected for this pilot application because of (1) its hyperlocal focus and descriptions of neighbourhood life in their reporting and (2) the availability of a large corpus of articles. At the time of data collection, most other neighbourhood newspapers in Hamburg covered smaller areas. In this sense, using local journalism as sociocultural data is exemplified using Elbe Wochenblatt as a paradigmatic case (Flyvbjerg 2006: 232). This study does not aim to discuss the media practices of the newspaper beyond a necessary contextualisation. 3.2 NATURAL LANGUAGE PROCESSING (NLP) OF LOCAL JOURNALISM As unstructured data and a by-product of economic and social activities, local newspaper articles require significant and meticulous preparation to provide tangible benefits to urban decisionmakers as sociocultural data. The preparation and curation of the unstructured data draw on the practices of the emerging field of NLP, advancing its use in data-based urban governance (e.g. Cai 2021). NLP is an algorithmically enabled analytical practice that combines linguistics, computer science and artificial intelligence to ‘structure large volumes of unstructured data’ (Cai 2021: 1). This study uses two primary methods of NLP to transform unstructured newspaper data into a source of subjective sociocultural urban data: geo-parsing and topic modelling. 3.2.1 Geo-parsing The geo-parsing process allows the extraction of location names and coordinates from the newspaper articles’ unstructured textual data. The ‘parsing’ process separated texts into a computer-readable format to detect terms that can be linked to geographical identifiers. In other words, geo-parsing involves two subprocesses that first extract location identifiers (i.e. place names) and then geocodes them (Wang et al. 2020). First, toponym recognition and geotagging identify names of geographical locations in the text based on name entity recognition. This task locates and classifies ‘named entities’ in unstructured text into predefined categories (e.g. person names, organisations, locations, time expressions, quantities, monetary values and percentages). After comparing multiple name entity recognition classification libraries, this study used spaCy, an open-source software library available for over 70 languages.3 Second, this study uses Nominatim4—an open-source geocoder to convert place names into geocoordinates— to geocode the geographical entities identified in the previous step. The algorithm builds a statistical model estimating the probability of a named entity referring specific geolocated place (Middleton et al. 2018). Entries with a score below a probability of 0.95 were checked manually. The geocoding of the 3511 newspaper articles identified 10,320 mentions of 1718 places in Hamburg. After the geo-parsing process, the database of newspaper articles required further preparation and data cleaning. A significant portion of place mentions refers to entities linked to areas across multiple neighbourhoods (i.e. districts, regions or Hamburg as a whole). As these place names are somewhat arbitrarily geocoded to a point within the area they represent, these place mentions clutter neighbourhood-level analysis with information related to areas outside the neighbourhoods.5 For this reason, the study excludes all mentions of place names referring to geographical areas above the neighbourhood level. This step excludes 1665 mentions of ‘Hamburg’, 356 mentions of the general parts of Hamburg (e.g. West Hamburg, South Hamburg) and 845 mentions of one of the seven districts. This filtering reduces the database to 7903 mentions of 1689 places. 376Mello Rose and Chang Buildings and Cities DOI: 10.5334/bc.300 Elbe Wochenblatt is distributed in an area that covers the entire west and south of Hamburg, including 34 of 104 official neighbourhoods (Figure 1). This area accounts for 41% of Hamburg’s total area and 35.3% of Hamburg’s population. The neighbourhoods serviced by the newspaper are highly diverse across various metrics ranging from population density to employment. For instance, the area of delivery includes the three wealthiest and the three poorest neighbourhoods of the city. The geocoding confirms the hyperlocal focus of the newspaper’s reporting, as over 89.5% of all place mentions in the database are in the delivery zone. Moreover, the vast majority (2539 articles, or 72.3%) of all 3511 analysed articles mention at least one place that can be located within a neighbourhood where the newspaper’s print edition is delivered. Therefore, the following study focuses on the areas in which the newspaper’s print version is distributed. The ‘cleaned’ corpus of newspaper articles thus includes 2539 articles mentioning 1424 different urban places in the delivery area a total of 6665 times. Tests of the text corpus and the parsed and geocoded places indicate that the number of times places in each neighbourhood are mentioned significantly correlates to that neighbourhood’s population (Pearson correlation = 0.667; p < 0.01) and area (Pearson correlation = 0.546; p < 0.01). No significant statistical relationship exists between the number of place mentions in an area and the area’s median income, housing prices or other tested socioeconomic factors (i.e. share of migrants, share of higher education degrees). 3.2.2 Topic modelling Topic modelling is an important step in NLP that allows the identification of different topics and topic distributions in large corpora of text. While there are several topic modelling methods, the most popular are latent dirichlet allocation (LDA) (Blei et al. 2003) and structured topic modelling (STM) (Roberts et al. 2019). The process follows the general procedure of clustering co-occurring lists of keywords into a predefined number of topics in the first step without needing any prelabelled data. To improve the certainty of identifying the correct topic, this study used both methods to identify 10 topics (i.e. co-occurring lists of keywords) in the text corpus of local newspaper articles. While both major topic modelling methods produce sensible results, a close analysis of the resulting 10 keyword lists forming each topic suggested the use of STM, as it allows the inclusion of additional Figure 1: Spatial distribution of the place mentions in the local newspaper. 377Mello Rose and Chang Buildings and Cities DOI: 10.5334/bc.300 information. In the case of this study, sentiment scores regarding the article’s style and tone were added to the analysis. These sentiment scores rest on ‘TextBlobDE’6 and the ‘Valence Aware Dictionary and Sentiment Reasoner’.7 In the second step, the topic modelling algorithm calculates the likelihood of an individual article being linked to each of the 10 topics (i.e. ‘topic score’). This analysis was based on the entirety of each article. As the sum of the 10 topic scores is equal to 1, the articles having a topic score > 0.5 (more than all other nine topics combined) in any given topic are considered to be significantly related to that topic. Articles without a topic score > 0.5 are understood not to be significantly linked to any of the 10 topics (e.g. combining multiple topics). Table 2 shows the topics identified in the topic modelling, the number of articles (with a respective topic score of at least 0.5) and the number of places mentioned in articles concerned with that topic. The number of place mentions varies significantly across different topics. For example, while topic 1, ‘pandemic’, was the most frequent topic in the corpus, articles clearly linked to topic 4, ‘mobility’, have the most place mentions. Due to the high threshold to be considered related to a particular topic, about 45% of articles are not linked to any topic. 3.3 EXEMPLAR USE CASES OF SOURCING LOCAL JOURNALISM FOR URBAN GOVERNANCE The following three use cases illustrate how the application of NLP to local newspaper articles can generate data to better comprehend a given area’s social and cultural fabric. While the first use case focuses on geo-parsing, the second demonstrates the application of topic modelling ID TOPIC LABEL ARTICLES PLACE MENTIONS KEYWORDS IN ORDER OF LIKELIHOOD (ROOT WORDS) 1 Pandemic 325 416 corona; vaccine; virus; test; pool; infect; centre; offer; hygiene; ship; trip; number; mask; pandemic; distance 2 Health and social care 275 1,318 health; hospital; help; also; care; work; homeless; employee; office; need; support; service; centre; contact; will 3 Mobility 233 1,515 traffic; bridge; park; station; railway; construct; car; road; plan; street; tree; residence; work; area; can 4 Sports 231 960 team; club; sport; game; league; will; football; championship; player; play; time; coach; goal; point; start 5 Politics 223 638 district; green; SPD; office; CDU; citizen; member; elect; initial; board; federal; politic; association; parliament; left 6 Schools and education 183 531 school; student; children; work; young; elementary; project; class; lesson; parent; teacher; educate; district; young; learn 7 Housing 151 391 build; new; plan; area; district; house; tenant; develop; rent; apartment; centre; euro; property; construct; office 8 Concerts and events 134 385 music; ticket; artist; theatre; perform; stage; artist; play; concert; show; band; musician; film; choir 9 Portraits of neighbourhood figures 93 277 like; father; photo; time; wife; friend; remember; even; still; just; now; always; live; want; know 10 Civic life 79 234 festival; museum; o’clock; children; offer; café; organ; church; event; open; market; culture; place; visitor; take 0No clear topic 1,588 2,671 Table 2: Results of the topic modelling. Note: CDU = Christian Democratic Union (Christlich Demokratische Union); SPD = Social Democratic Party (Sozialdemokratische Partei). 384Mello Rose and Chang Buildings and Cities DOI: 10.5334/bc.300 Gentzkow, M., & Shapiro, J. M. (2010). What drives media slant? Evidence from U.S. daily newspapers. Econometrica, 78(1), 35–71. DOI: https://doi.org/10.3982/ECTA7195 Ghermandi, A., & Sinclair, M. (2019). Passive crowdsourcing of social media in environmental research: A systematic map. Global Environmental Change, 55, 36–47. DOI: https://doi.org/10.1016/j. gloenvcha.2019.02.003 Gregory, I. N., & Paterson, L. L. (2020). English language and history: Geographical representations of poverty in historical newspapers. In The Routledge handbook of English language and digital humanities (pp. 418–439). Routledge. DOI: https://doi.org/10.4324/9781003031758-22 Institut für Medienund Kommunikationspolitik. (2022). Ranking—Die zehn größten deutschen Medienkonzerne 2021 (June). Statista. https://de.statista.com/statistik/daten/studie/194686/umfrage/ die-10-groessten-medienkonzerne-in-deutschland/ Jacobs, J. (1992). The death and life of great American cities [1961]. Vintage/Penguin Random House. Jahedi, S., & Méndez, F. (2014). On the advantages and disadvantages of subjective measures. Journal of Economic Behavior & Organization, 98, 97–114. DOI: https://doi.org/10.1016/j.jebo.2013.12.016 Kahneman, D., & Krueger, A. B. (2006). Developments in the measurement of subjective well-being. Journal of Economic Perspectives, 20(1), 3–24. DOI: https://doi.org/10.1257/089533006776526030 Kitchin, R. (2014). Big data, new epistemologies and paradigm shifts. Big Data & Society, 1(1), 2053951714528481. DOI: https://doi.org/10.1177/2053951714528481 Kitchin, R., & McArdle, G. (2016). What makes big data, big data? Exploring the ontological characteristics of 26 datasets. Big Data & Society, 3(1), 2053951716631130. DOI: https://doi. org/10.1177/2053951716631130 Kitchin, R., & McArdle, G. (2017). Urban data and city dashboards: Six key issues. In Data and the city (Vol. 1). Routledge. DOI: https://doi.org/10.4324/9781315407388-9 Lomborg, S., Dencik, L., & Moe, H. (2020). Methods for datafication, datafication of methods: Introduction to the Special Issue. European Journal of Communication, 35(3), 203–212. DOI: https://doi. org/10.1177/0267323120922045 Macků, K., Caha, J., Pászto, V., & Tuček, P. (2020). Subjective or objective? How objective measures relate to subjective life satisfaction in Europe. ISPRS International Journal of Geo-Information, 9(5), art. 5. DOI: https://doi.org/10.3390/ijgi9050320 Maffei, S., Leoni, F., & Villari, B. (2020). Data-driven anticipatory governance. Emerging scenarios in data for policy practices. Policy Design and Practice, 3(2), 123–134. DOI: https://doi.org/10.1080/25741292.2020. 1763896 Mejias, U. A., & Couldry, N. (2019). Datafication. Internet Policy Review, 8(4). https://policyreview.info/ concepts/datafication. DOI: https://doi.org/10.14763/2019.4.1428 Middleton, S. E., Kordopatis-Zilos, G., Papadopoulos, S., & Kompatsiaris, Y. (2018). Location extraction from social media: Geoparsing, location disambiguation, and geotagging. ACM Transactions on Information Systems, 36(4), 40:1–40:27. DOI: https://doi.org/10.1145/3202662 Ozgun, B., & Broekel, T. (2021). The geography of innovation and technology news—An empirical study of the German news media. Technological Forecasting and Social Change, 167, 120692. DOI: https://doi. org/10.1016/j.techfore.2021.120692 Paterson, L. L. (2020). Mapping austerity: Geographical text analysis of UK place-names in The Guardian and The Daily Telegraph. In Multimodal approaches to media discourses. Routledge. DOI: https://doi. org/10.4324/9780367332907-4 Porter, C., Atkinson, P., & Gregory, I. (2015). Geographical text analysis: A new approach to understanding nineteenth-century mortality. Health & Place, 36, 25–34. DOI: https://doi.org/10.1016/j. healthplace.2015.08.010 Rettberg, J. W. (2020). Situated data analysis: A new method for analysing encoded power relationships in social media platforms and apps. Humanities and Social Sciences Communications, 7(1), art. 1. DOI: https://doi.org/10.1057/s41599-020-0495-3 Rieder, G., & Simon, J. (2016). Datatrust: Or, the political quest for numerical evidence and the epistemologies of big data. Big Data & Society, 3(1), 2053951716649398. DOI: https://doi. org/10.1177/2053951716649398 Roberts, M. E., Stewart, B. M., & Tingley, D. (2019). stm: An R Package for structural topic models. Journal of Statistical Software, 91, 1–40. DOI: https://doi.org/10.18637/jss.v091.i02 Sarın, P., & Uluğtekin, N. (2019). Analyzing newspaper maps for earthquake news through cartographic approach. ISPRS International Journal of Geo-Information, 8(5), art. 5. DOI: https://doi.org/10.3390/ ijgi8050235 Scheufele, D. (1999). Framing as a theory of media effects. Journal of Communication, 49(1), 103–122. DOI: https://doi.org/10.1111/j.1460-2466.1999.tb02784.x 385Mello Rose and Chang Buildings and Cities DOI: 10.5334/bc.300 TO CITE THIS ARTICLE: Mello Rose, F., & Chang, J. (2023). Urban data: harnessing subjective sociocultural data from local newspapers. Buildings and Cities, 4(1), pp. 369–385. DOI: https://doi.org/10.5334/bc.300 Submitted: 10 February 2023 Accepted: 05 June 2023 Published: 30 June 2023 COPYRIGHT: © 2023 The Author(s). This is an open-access article distributed under the terms of the Creative Commons Attribution 4.0 International License (CC-BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. See http://creativecommons.org/ licenses/by/4.0/. Buildings and Cities is a peerreviewed open access journal published by Ubiquity Press. Swyngedouw, E. (2005). Governance innovation and the citizen: The Janus face of governance-beyond-thestate. Urban Studies, 42(11), 1991–2006. DOI: https://doi.org/10.1080/00420980500279869 Taylor, L., & Richter, C. (2015). Big data and urban governance. In J. Gupta, K. Pfeffer, H. Verrest, & M. Ros-Tonen (Eds.), Geographies of urban governance (pp. 175–191). Springer. DOI: https://doi. org/10.1007/978-3-319-21272-2 Valdez, A.-M., Cook, M., & Potter, S. (2018). Roadmaps to utopia: Tales of the smart city. Urban Studies, 55(15), 3385–3403. DOI: https://doi.org/10.1177/0042098017747857 Voukelatou, V., Gabrielli, L., Miliou, I., Cresci, S., Sharma, R., Tesconi, M., & Pappalardo, L. (2021). Measuring objective and subjective well-being: Dimensions and data sources. International Journal of Data Science and Analytics, 11(4), 279–309. DOI: https://doi.org/10.1007/s41060-020-00224-2 Wang, J., Hu, Y., & Joseph, K. (2020). NeuroTPR: A neuro-net toponym recognition model for extracting locations from social media messages. Transactions in GIS, 24(3), 719–735. DOI: https://doi.org/10.1111/ tgis.12627 Zuboff, S. (2019). The age of surveillance capitalism: The fight for a human future at the new frontier of power. PublicAffairs.