Skip to main content
SearchLoginLogin or Signup

From exceptional architecture to city icons? Analyzing data scraped from Flickr

Published onJun 20, 2022
From exceptional architecture to city icons? Analyzing data scraped from Flickr
·

Abstract

The link between the online circulation of photographs and the iconicity of an exceptional architectural project remains empirically under-researched. Similarly, so is the transformative effect that such circulation might have on the image of a city. This paper provides a proof of concept about what large volumes of unstructured data, scrapped from the photo sharing platform Flickr, organized and categorized by automated computer scripts, can tell us about the performance of an exceptional architectural project as a city icon. Using the case study of the Elbphilharmonie in the city of Hamburg, Germany, we present findings of an analysis of photographs and their corresponding metadata – hashtags captions and comments – extracted from Flickr using the keyword “Hamburg”. The objective of the analysis is to identify the effects that the Elbphilharmonie as an exceptional architecture had on the circulation of photographs of Hamburg. Does the Elbphilharmonie manifest itself as a landmark for Hamburg? Does attention move away from the traditional point of interest of the Alster lake and down to the river and port as intended by the proponents of the project? The paper describes the incremental quantitative and qualitative shift of attention towards the port of Hamburg captured in the Flickr data. Indications of the viral effects of the Elbphilharmonie lie in compounding the findings of the topic clustering, ranking of hashtags, the volumes of photographs taken from locations that aim to replicate the official photograph, and the heterogeneous nature of the posts. We conclude by pointing to the limitations of such research methods. Data scraped from the Flickr platform can generate findings pertaining to the performance of exceptional architecture as city icons. Nevertheless, care must be taken to differentiate between popular places that are consumed and symbolic places that are revered. Keywords: Flickr, star architecture, icons, city identity, urban development, exceptional architecture, social media data

Introduction

Exceptional architecture is often commissioned with the deliberate and declared objective that it contributes to transforming the city’s form and users’ experiences of the city. From the mid-2000s, with the progressive and simultaneous use of handheld smart devices and online social media platforms, the commissioning of public cultural facilities as exceptional, highly recognizable architecture, otherwise known as “star architecture,” has intensified in cities, particularly those cities that were seeking to emulate the so-called Bilbao effect (Ponzini and Akhavan 2020; Plaza et al. 2015; Gravari-Barbas, Renard-Delautre, and Oakman 2015; González 2011). A common unsubstantiated assumption in most scholarly discussions of star architecture is that the iconicity of exceptional architecture contributes to achieving the intended shifts in scale, spatiality, and content of attention in a city (e.g., Lindsay and Sawyer [2021] and Patterson [2020]). Yet iconicity is a theoretical construct, developed by sociologists (e.g., Sklair [2017]) and architectural theorists and historians (e.g., Jencks [2011]) based on observation. The link between the online circulation of large volumes of photographs of an exceptional architectural project and the iconicity of such a project remains empirically underresearched, particularly how such online circulation not only reflects the status of such a project as a city icon and symbolic landmark but also supports achieving such status. Equally underresearched is the transformative effects that such circulation might have on the image of a city. These research gaps are due to methodological challenges as well as lack of relevant research skills that limit the capacity of architecture scholars to undertake such investigations on their own. Based on the idea that Big Data can provide insights into aspects of human behavior that have previously been difficult to observe (Einav and Levin 2013), this article presents the findings of a collaboration between scholars of architecture and informatics.

The aim of this article is to show the potentials and limitations of the analysis and interpretation of large volumes of unstructured data scraped from the photo-sharing platform Flickr and then organized and categorized by automated computer scripts. The objective is to present a proof of concept of what the analysis of online photographs and texts can tell us about the performance of an exceptional architectural project as a city icon. The step change in the scale and scope of the sources of data (Taylor, Schroeder, and Meyer 2014) enable new ways of thinking and lead researchers to ask new questions “born from the data” (Kitchin 2014). The premise of the article is that the millions of photographs taken by individuals of cities via handheld devices that are then circulated on the internet can provide new insights into exceptional architectural projects, namely, how such projects that are planned top down as transformative agents in cities are actually received and perceived from below, that is, from the user perspective. Our hypothesis is that since hashtagged photographs of a city and the associated metadata capture the development of areas in a city receiving attention or trending, analysis of such photographs and text data can provide insights that support identifying the role that exceptional architecture plays in therein. There are limitations to the conclusions that can be drawn from such findings, discussed hereafter. The most pertinent limitation pertains to the representativeness of social media users in general and a certain profile of users of our selected photo-sharing platform Flickr in particular. Findings, therefore, offer only a glimpse into a certain universe of users, and caution must be taken about generalizing. Cross-referencing such findings with an analysis of newspaper articles and another image database, such as Getty Images, can sharpen these findings. “Ground-truthing findings through site visits and interviews” (Williams 2020, x) can enhance the validity of drawn conclusions. This would support evaluating how a city has steered its urban development and how future investments may benefit from social-media-based tactics.

Using the case study of Hamburg and its iconic concert hall, the Elbphilharmonie, this article tells the story of the evolution of the image of the City of Hamburg in relation to the Elbphilharmonie as captured by Flickr users. The quantitative analysis identifies volumes of information that required qualitative analysis, which in themselves have limited yield but, when considered in relation to each other, are demonstrative. Heat maps and classic topic analysis were used in an iterative process to analyze and interpret the data. Heat maps that visualize the geographic distribution of density of photographs uploaded in 2009, 2014, and 2019 show the development of Hamburg’s presence on Flickr and point to new and sustained attention to the Elbphilharmonie and the harbor in addition to the popular Alster Lakes. The emergence of an intense hotspot in the heat map of 2019 at the location of the Elbphilharmonie, as well as the increased volume and intensity of photographs on the surrounding quays, demonstrate the shift from the city center to the harbor areas in Hamburg. Hotspots along the banks of the River Elbe, particularly in positions that facilitate the replication of the iconic photographs of the Elbphilharmonie, are indicative of the viral effects of its exceptional architecture. The hashtag analysis shows that the hashtags “alster” and “altona” have reducing representation in photographs. Finally, the topic analysis reveals the “hafen” topic as the dominant topic across all three years. Together, these findings support the indication of the diversification of attention in the city from the Alster Lakes toward the banks of the Elbe in the vicinity of the harbor. The hashtags listed with photographs describe the Elbphilharmonie’s association with historical points of interest and unintended associations with particular actors and events. The Elbphilharmonie, and other cultural points of interest, compete with influencers and popular commercial offerings in the city. These are not surprising findings, yet they deliver empirical evidence about how the data on an online photo-sharing platform captures the spatial shift of attention toward exceptional architecture.

Exceptional architecture as iconic architecture with viral effects

When public actors commission architecture projects, these commissions are always linked to public actors’ desire to support political projects. Especially at critical junctures in the development of a collectivity, architecture is mobilized by public actors to construct cultural forms that support political projects by foregrounding certain cognitive constructs. Vale (1992) elaborates on how the production of landmark buildings contributes to nation building through the communication and consolidation of state power. The accentuation of globalization as an economic, cultural, and political project destabilized the relevance of some of the established landmarks and kickstarted efforts to establish visual representations of new identities. This happened simultaneously with the progressive use of handheld smart devices and social media platforms and accentuated circulation of photographs on the internet from the mid-2000s. Public actors and agencies sought to reposition their cities on global networks by commissioning internationally acclaimed architects for the design of public cultural facilities (Thierstein, Alaily-Mattar, and Dreher 2020). These express global cultural references and, when photographed and hashtagged, draw global attention and recognition as an “electronic word of mouth” (Gunawan and Huarng 2015, 2237). Global attention and recognition also serve the construction of local identities and stabilization of meanings. The worldwide circulation of digital photographs of new symbolic landmarks in cities is a key driver of a city’s image at multiple scales.

Sklair (2017, 17-18) notes that the circulation of photographs is central in the production and iteration of iconicity. The photographs become representative of lifestyles, meanings, and values that do not necessarily relate to the program of the building itself (Jencks 2006). “Icons are spectacular structures that we visit in droves, expecting to be transformed by the experience, that it will leave us seeing the world a little differently” (Khan 2018); a post and a photograph on social media capture data about this moment. In this article, we are interested in what such data can tell us about the iconic status of exceptional architecture and its viral power, explored herein as the multiplication of the iconic photograph. In line with the shift in consumerism from collecting things to collecting experiences (Klingmann 2007), the proliferation of photographs by locals or tourists with an iconic built artifact on social media draws attention to the broader potential and risks of the development of “instagrammable architecture” (Fiocco and Pistone 2020; FitzGerald 2020) and a city’s tactics toward consumer behavior and viral marketing (Gunawan and Huarng 2015), which fall outside of the scope of this article.

The iconic status of built artifacts and their photographs is not given but rather derived from circulation. Plaza et al. (2015) argue that the success of the Guggenheim Museum in Bilbao (GMB) in attracting media attention should be mostly attributed to the coincidence of the opening of the GMB in 1997 with the launch of the internet in 1995. The speed, volumes, and, particularly pertinent to social media, the personalization—across the spectrum from the professional photographer to the individual “selfie”—and the geographical scope of the dissemination of user-generated photographs are useful indicators of such circulation. Sentiment is carried in the personalization in the text and photograph itself. “The public culture’s image has changed as citizens present another perspective which reveals more truth and objectivity than traditional journalism and made its photographs iconic” (Lai 2011, 702). The narratives and photographs circulated by the producer/supply side of planning authorities and developers are offset by the demand/user side.

Data hosted on social and photo-sharing media platforms present a great potential for uncovering how exceptional architecture projects are received in cities and how this perception goes on to change the way cities are perceived. In the social sciences, the rush to use data from social media platforms resulted in a “data gold rush” (Kennedy et al. 2015, 1) and a “computational turn” (Felt 2016, 1). In the spatial sciences, there is an equal appeal for using such data for describing, measuring, and understanding spatial phenomena. Analyzing the city through data retrieved from location-based social networks has received considerable attention as a promising method for applied research (Marti, Serrano-Estrada, and Nolasco-Cirugeda 2019). Currid and Williams (2010) use geotagged images available through the Getty Images database to generate “buzz maps” of New York City and Los Angeles that help identify macro-geographical patterns. “The maps reveal that paparazzi, marketers and media—those not conventionally involved in city development—have unintentionally played a significant role in the establishment of buzz and desirability in city development” (Williams 2020, 120). Similarly, using georeferenced social media data to map the sensorial and emotional layers of cities, researchers at goodcitylife.org developed what they call “happy maps,” “chatty maps,” and “smelly maps” to offer an alternative cartographic representation of a city (see, for example, Quercia, Schifanella, and Aiello [2014[). Williams (2020, 120-21) argues that “if the urban theorist Kevin Lynch (1918–1984) were living today (…) he would have been obsessed with using the thoughts and ideas posted on social media to better understand the city—thereby creating our era’s version of cognitive maps (also called mental maps) (…) Creating a map from geo-registered social media data similarly notes people thoughts in the place where they thought them, which becomes a landscape of feelings, emotions and personal landmarks.”

The use of such data is not without its challenges, including ethical considerations, and has given rise to a stream of polemical arguments over the validity of this source of information. Boy and Uitermark (2017) caution that media practices on social media platforms are subject to a set of unwritten rules that, in Instagram induce a degree of conformity while capturing local scenes showcasing patronage of exclusive places aimed at boosting user’s status and inadvertently the status of those places too, as does Currid and Williams (2010) related to Getty Images of high-profile events. It is worthwhile to understand what and how the seemingly endless supply and circulation of amateur photographs that are posted, hashtagged, and shared contribute to the narratives of power and to the political project behind exceptional architecture projects, using an online platform that showcases these places. 

The Hamburg Elbphilharmonie: A game changer for Hamburg?

The Elbphilharmonie in Hamburg, Germany is an example of an exceptional architecture project that was commissioned by the public sector to serve as a new landmark for a city and to signal the start of a new chapter in its history (Figure 1). The building was designed by Pritzker Prize recipients Herzog & de Meuron.

Figure 1

Press photographs of the Elbphilharmonie, which can be replicated from the Überseebrücken and Theater an der Elbe. Source: Thies Rätzke 2021 (left) and Maxim Schultz 2018 (right).

The inception of the idea of this project goes back to the early 2000s, when the City of Hamburg was undertaking a large urban transformation project along its former harbor district. The Free and Hanseatic City of Hamburg is a city-state in Germany, located Southeast of the River Elbe's estuary mouth on the North Sea at the junction of the River Elbe with the Rivers Alster and Bille (Figure 2). Hamburg’s ancient river harbor enabled the city to become an important trading center on a regional and continental scale. Over time, the harbor of Hamburg became the largest port in Germany, and the fate of the city has been linked to the economic significance of its port. For centuries, the City of Hamburg expanded north of the harbor. Throughout the twentieth century, the Alster Lakes served as important identity marker for the city with its recreational area in the heart of Hamburg and whose edges are lined by residential and commercial buildings with white facades and copper-clad roofs, with the most famous one being the Atlantic Hotel. In the 1990s, the Hamburg Senate decided to repurpose its warehouse district, where most of the land is owned by the City of Hamburg through the harbor company. This district had been abandoned with the advancement of containerization. Through a city-owned subsidiary, an urban transformation project called “HafenCity” set out to transform a 157-hectare former harbor land area into a mixed-use urban district. The district itself would later be called “HafenCity.” The Elbphilharmonie was constructed as landmark at a strategic site in HafenCity. The building was accompanied by an initial public euphoria (Balke, Reuber, and Wood 2018) fueled by local media exposure (for a content analysis of local newspaper articles of the Elbphilharmonie inception phase, refer to Alaily-Mattar, Akhavan, and Hein [2021]). In October 2005, the Hamburg State Parliament consensually approved building the Elbphilharmonie and authorized the Hamburg Senate to award the project. Construction commenced in 2007; after juridical struggles that resulted in a temporary construction standstill were resolved eventually in July 2013 (Lauber 2016), the Elbphilharmonie project was completed and hosted its first concert in January 2017. The final price tag was €866 million—on the evolution of the costs, refer to Zinnecker (2019).

Figure 2

Map of Hamburg (2019) showing the Port of Hamburg area, HafenCity, and points of interest. Source: Authors‘ own elaboration, contains information from OpenStreetMap and OpenStreetMap Foundation, which is made available under the Open Database License 2021.

The inception of the HafenCity project as a mixed-use development in the mid-1990s indicated a commitment of the City of Hamburg to transform its port area. The naming of the project “HafenCity,” translated “Port City,” signaled a strong recognition of the importance of sustaining the idea of the port in the imagination of the city. With this project, the city intended to re-establish the connection between the River Elbe and the city center, giving Hamburg a new direction for growth, down to and along the river (Schubert 2014). A newspaper article published in 2002 in a German newspaper, titled “Finally, Hamburg becomes PortCity” (authors’ translation), notes that for 1,000 years, the city had expanded northward and, as it were, turned its back on its river; the HafenCity project marks a new chapter in the history of Hamburg, in that same area where the city's history began (Iken 2002). The city’s renewed interest in the port was a shift of focus of urban planning in Hamburg from one that revolved around the Alster Lakes to one that expanded the city to the River Elbe. Including the name of the river in the name of the project further confirms the city’s commitment at the time of the shift to the river. Indeed, the Elbphilharmonie was communicated to be “a symbol of the expansion of downtown Hamburg to the south into the port areas on the banks of the Elbe (Göppert 2005). The quest to build an internationally recognized architectural symbol was framed as being not only in the spirit of Hamburg’s maritime internationalism but also a necessity for staying competitive on an international scale by signaling to the world that the City of Hamburg no longer distances itself from its port but rather endorses it. A lot of emphasis was placed on communicating the idea that the symbolic value of the Elbphilharmonie manifests itself not only in its landmark value but also in terms of its sense-making value. One newspaper article explains this as follows: “A James Bond film was shot in Hamburg a few years ago. The first question the director asked was where in the city is the building by which the world can tell that we are in Hamburg. The director could not be helped and he chose the Atlantic Hotel. We now know how important the Guggenheim Museum in Bilbao or the Opera in Sydney are for a metropolis. A Philharmonie could play this role in Hamburg” (Schirg and Iken 2005). Much of the legitimization for the project was built around the logic that Hamburg needed a landmark. Allegories were drawn to the Sydney Opera House and the Eiffel Tower.

Against this background, it is then valid to ask whether the Elbphilharmonie met this objective. What effect did this exceptional architecture project have on the circulation of photographs of Hamburg? Did the Elbphilharmonie manifest itself as a landmark for Hamburg? Did attention move away from Alster Lakes and down to the river and port? Did the port begin to reoccupy a more significant place in the imagination of the people? Most importantly, what research methods can we apply to investigate such questions?

Methodology

It is possible to retrieve data that is relevant to our research questions via website scraping with or without an application programming interface (API) to the official provider. We decided to restrict our research exercise to Flickr, as the Flickr platform was more accessible and robust due to an official API, unlike Instagram, which lacks such an official API and has ever-changing protocols. Furthermore, the geolocation is more accurate in Flickr than Instagram due to the API filter applied in the latter. As such, Instagram posts are not directly georeferenced to their exact location (Marti, Serrano-Estrada, and Nolasco-Cirugeda 2019) but rather refer to a so-called location ID, which incorporates all photographs close to a specified landmark. This decision places limitations on the interpretation of the findings due to the profile of Flickr users who are photographers passionate about connecting with others through photography (Grenwetzki 2018). Nevertheless, the objective of the article is to present a proof of concept, and further research can apply this methodology using data extracted from other photo-sharing platforms. We extracted data from Flickr using a search on the hashtag “hamburg,” specifically for the years 2009, 2014, and 2019. These dates describe five-year intervals from the last summer before the 2020 COVID-19 pandemic. The tool that we developed and used for obtaining our data sets is a Python Flask web application with a PostgreSQL database running behind a NginX reverse proxy. Our tool, called “Archistar,” contains its own user management and privilege system. Its functionalities allow the scraping of photographs from Flickr for a specific search tag (e.g., Hamburg). Initially, we implemented the request to the Flickr API, which would return all photographs for a defined bounding box of longitudes and latitudes; however, due to nondeterministic responses from Flickr, we discarded using this option and scraped all our data based on the search tag “Hamburg.’” As the default setting, we requested only photographs that are geotagged be returned. The exact API call we used was flickr.photos.search1. Set parameters are the api_key, tags=[‘Hamburg’], min_upload_date, max_upload_date and has_geo=1. We decided to use the settings for upload_date instead of taken_date under the assumption that both dates lie close together and because a taken_date is not available for all photographs. Since Flickr restricts the use of every API key to approximately 3,600 requests per hour, we scraped all our data sets month by month to avoid reaching this limit. This can be done by setting the time interval accordingly before starting a scrape. Afterward, we combined all the data from 12 months to a unified yearly data set using the onboard functionalities of our tool. We determined the associate of a photograph to a specific year, if its UNIX time stamp corresponded to our target date interval. We downloaded all the data of a data set in a Comma Separated Value (CSV) file format to a local machine for further preprocessing (Table 1).

Since we are interested in the shifts over time, a five-year interval related to before, during, and after construction of the Elbphilharmonie was deemed sufficient for the purposes of this study. The consideration of the total number of photographs that we anticipated collecting also played a role in selecting fewer rather than more years of data. This was due both to the onerous time it takes to extract data, at approximately six hours of computing to extract one year’s data, and the storage and processing capacity for tens of thousands of photographs. Data mining is limited differently by the various platforms to avoid the extraction of large amounts of data; we put much effort into respecting these limitations, applying ethical practices by retrieving only public data and applying an anonymization process that pixelates faces in photographs. Much of the effort in designing the tool is a consequence of these limitations. This includes limiting the amount of data scraped for a given request (i.e., how many times a request will be accepted by the platform in a specified amount of time). As a result of the collection of data in a piecemeal process, scripts were written that were able to compare different data sets in time intervals and then merge them while removing duplications. Despite efforts to extract unbiased data, we came to recognize that we cannot absolutely guarantee the absence of hidden biases. Due to API parameters that were not explicitly explained, we encountered a lack of transparency that could have an unwanted effect on the returned data. For instance, Flickr has accuracy levels for photograph searches according to geolocation, but Flickr does not make explicit what the criteria for the different levels are and how it alters the API response.

Once we collected the data for each month of each year, we compiled a consolidated data set of all three years (Table 1). We undertook an initial classic topic analysis of the texts associated with the photographs (i.e., descriptions, hashtags, captions). Using scripts, we filtered the data (e.g., identifying all airplane hashtags) and produced an infographic of the main topics that allows users to flexibly explore topic–term relationships using relevance. The subtopics within topics are also identifiable. The topic analysis was done using nonnegative matrix factorization (NMF)2. In order to process the full text, we vectorized it via a commonly used vectorizing technique called “Term Frequency-Inverse Document Frequency” (TF-IDF). To compute the TF-IDF vectors from our data, we apply a CountVectorizer3 following an L1-normalized TfidTransformer4. These document vectors can be merged into a document–word matrix, where every row represents a document and every column represents a single word. NMF uses this document–word matrix to factorize it into a document–topic and a topic–word matrix. These matrices have a lower hidden dimension than the original matrix. With this, NMF tries to represent the high-dimensional data, produced with TF-IDF vectorization, in a latent, lower-dimensional space. We can interpret this latent space as a distribution over underlying topics (i.e., documents that share a common topic will be closer together in this space than other documents). The document–topic matrix describes to which topics each of the document contributes, while the topic–word matrix shows which words construct a topic. The two latent space matrices are generated by an alternating optimization algorithm (Cichoki and Phan 2009). An explorative and iterative approach to the topic analysis was adopted, including applying the automated method, reviewing the results, spatially visualizing the results, and using these to interpret the data, specifically seeking out topics or semantics that were deemed relevant to the research questions and interrogating these assumptions for plausibility (Sievert and Shirley 2014). The merging of other words or themes that are associated with one another, such as “elbphilharmonie” with “elphi,” “concerthall,” “philharmonic,” “elbephilharmonic,” “elfi,” “elbphilharmonic,” “konzertsaal,” “konzerthaus,” “elbphildisharmonie,” “elbephilharmonicconcerthall,” and “elbfilharmonie,” was done by scripts.

Table 1

Data points

2009

2014

2019

Total

All photographs that carry the hashtag “Hamburg,” including those without geolocation

142,852

133,089

74,578

350,519

All photographs that carry the hashtag “Hamburg,” with geolocation

22,485

20,078

15,843

58,403

All photographs that carry the hashtag “Hamburg,” with a geolocation in Hamburg(1)

21,479

19,068

13,649

54,196

Unique posts that carry the hashtag “Hamburg” and with a geolocation in Hamburg(1)

8,883

11,414

9,108

29,405

All photographs with the “*el*phi*” hashtag

135

424

417

976

Unique posts with the “*el*phi*” hashtag

50

332

367

749

  1. We delineated Hamburg using the geotag of Hamburg with the latitude 9.9937 and longitude 53.5511. According to hamburg.de/info/3277402/hamburg-in-zahlen/ the radius of the city is circa 20 km. Therefore, the difference in the latitude and longitude values of the posts in the City of Hamburg must be <= 0.15.

Table 1: Overview of the size of the data sets scraped from Flickr; the grey highlighted row is the data set that we used for our analysis. Source: Authors’ own elaboration.

The spatial visualizations of the data set were achieved by importing the data set in CSV file format into QGIS (an open source application) with which point location and heat maps were created and locational queries on the data were made. A heat map for each of the three years was produced applying a built-in feature in QGIS called “GIS core density estimation.” Several filters were then applied, allowing different data sets to be visualized in maps. Geolocated photographs were represented on a map both as individual points and by applying the heat map tool. We used topics to identify which photographs were irrelevant to our research question. We used heat maps to identify particular areas of interest and then ran topic analyses to understand why particular hotspots exist. It was useful to use these two approaches iteratively and in different directions. The iterative visualization of data clustered according to topic as heatmaps and vice versa enabled a thorough structuring of the unorganized data. For example, as a result of mapping of the “airplane” topic datapoints, the outlying and isolated character of the data was revealed and allowed the subset to be set aside and thus reduce so-called noise. It became clear that in order to explain the existence of hotspots in heatmaps, the photograph and topic or text should be considered in relation to each other. For example, it revealed that one main topic related to a nightclub called “juiceclub” and that multitudes of photographs (N = 1,912) were uploaded against one Flickr post. By merging photographs uploaded as a batch or gallery into one post but keeping their unique identity numbers, the number of data points reduced substantially (N = 47) and the respective hotspot reduced proportionally on the heat map. Since photographs in this particular topic are assumed to be irrelevant to this study, we did not further investigate each photograph. To elaborate, what appears as one point in the map is actually 1,912 photographs placed over one another in 47 posts; in this instance, by reducing this to unique posts, the hotspot loses its intensity relative to other unique posts in the study area (Figure 3). The automated mining and cleaning of data is necessary when we consider that 750 unique posts of the 58,403 photographs collected refer to the “elbphilharmonie,” or just 0.012 percent—an analogue process to reach the same result is simply inefficient.

Figure 3

Heat map 2009 showing the difference between all photographs (e.g., juiceclub N = 1,912) and unique posts (e.g., juiceclub N = 47) uploaded on Flickr. Source: Authors‘ own elaboration, contains information from OpenStreetMap and OpenStreetMap Foundation, which is made available under the Open Database License 2021.

Reading the shifting center of attention in the City of Hamburg using heat maps of photographs uploaded on Flickr

The heat maps of datapoints related to unique posts show that through these years, while some hotspots persisted from 2009 to 2019, new hotspots have emerged, particularly along the river, and some hotspots of established landmarks have lost intensity. At a city scale (Figure 4) in 2009, the hotspot the “juiceclub” in the Holsten district in the northwest (Hotspot A in Figure 4), in the Neustadt at Reeperbahn and Landungsbrücken (B), and Hamburg’s largest landing piers at St. Pauli district, are most prominent. Other hotspots are located at the location of landmarks in Altstadt (old town, (C)) including the Michel Tower—the colloquial name for the bell tower of St. Michaelis Church, the traditional landmark of Hamburg. By 2014, the opening of the Theater an der Elbe on the south banks of the River Norderelbe accessed via the Alte Elbtunnel and Landungsbrücken (D) by boat on the northern banks may be responsible for the increased photographs from there. The Altstadt hotspots are consistent, but further consolidation of the Speicherstadt hotspot (E) to the south of the Altstadt is evident, including the emergence of the Elbphilharmonie. In 2019, the “airplane”-related posts are prominent in the north at Hamburg Airport, as is made evident by hotspots in the vicinity of the airport and surrounds (F), as are the Speicherstadt hotspots (G). A new hotspot (H) emerges further east, namely the prizewinning Elbbrücken train stations designed by Gerkan, Marg, and Partner Architects, which was completed in December 2019. No notable difference is evident in the Alster Lakes stretching north. In the southern part of Hamburg, datapoints extend to the Wilhelmsburg district, the large southernmost portion of the harbor; however, these are insufficient to generate a hotspot on the map.

Figure 4

Heat maps of unique hashtagged photographs that carry the hashtag “Hamburg,” are geotagged and are uploaded on Flickr. Source: Authors own elaboration, contains information from OpenStreetMap and OpenStreetMap Foundation, which is made available under the Open Database License 2021.

A zoom-in of central Hamburg shows the changes in hotspots (Figure 5). The Miniatur Wunderland (9), a warehouse housing a large model railway running through miniaturized replicas characteristic of different countries and cities with their key landmarks, including Hamburg from 2002 and the HafenCity and Elbphilharmonie subsections from 2013, is a consistently prominent hotspot across the years. The intensity of the hotspot at the Michel Tower (5) of St. Michaelis remains almost relatively consistent from 2009 to 2019. But, comparatively, in 2019, it is less intense than the Elbphilharmonie (1).

Figure 5

Heatmap of central Hamburg. Source: Authors‘ own elaboration, contains information from OpenStreetMap and OpenStreetMap Foundation, which is made available under the Open Database License 2021.

The intensity of uploaded photos that are taken from the south bank of the River Norderelbe at the Theater an der Elbe (6) and on the north bank from Landungsbrücken (4) and spreading east along Elbpromenade (10), a 1.3 km walkway along the river designed by Zaha Hadid Architects, appear more dispersed from 2009 to 2014 with increasing prominence of the Überseebrücken (11), a covered pedestrian bridge connecting the shoreline to the quays, to the Baumwall metro stations (12)—the position from which the iconic photograph of the Elbphilharmonie is easily replicated—and culminating at the Poggenmühlebrücke (13) at the east end of the Speicherstadt district. Photographs on the water on boats also show increases. Striking new intense hotspots appear in 2019 at the location of the Elbphilharmonie (1), which was already evident and equivalent in intensity to the Michel Tower (5) in 2014 despite the building being under construction.

The fact that the Elbphilharmonie functions as concert hall also contributes to the increased intensity, as people revisit and take photographs repeatedly at different music events, rather than it being a once-off photo opportunity, as is typically the case for other cultural city landmarks. In addition, a hotspot emerges around the Elbbrücken train station (8). In contrast, the Atlantic Hotel (14) is not identifiable as a point of interest in the city despite its perceived prominence. Hotspots with marked decrease in intensity include the award-winning, ship-like Docklands Office (15) building designed by Bothe Richter Teherani Architects in the Altona Fischmarkt district. The building opened in 1995, offering a publicly accessible raised viewing platform facilitating spectacular views over the River Elbe; by 2019, such views are taken from the Elbphilharmonie’s viewing platform. The Internationales Maritime Museum (16) to the east of the Speicherstadt is only visible in 2014, while the area just south of it around HafenCity Hamburg University (17) gains in prominence. The hotspot at the St. Pauli metro station (18) in 2009 moves westward by 2014 along Reeperbahn (3), an entertainment boulevard, and further west by 2019 but loses intensity as it moves. By 2019, the intensity at the historical building at the Alter Elbetunnel (7) and surrounds is significantly less than around the Überseebrücken (11) and Baumwall metro stations (12). By 2019, the Elbpromenade hotspots (10) are less intense compared to the Speicherstadt hotspots.

An arc-shaped series of hotspots can be identified on the northern banks of the River Norderelbe between Landungsbrücken, along the Elbpromenade (10), and along the Zollkanal, the historical customs canal that separated the Altstadt (Old City) from the southern free port, to the Hamburg Hauptbahnhof, the main train station (19). The main station appears to be constant despite Flickr reaching its peak popularity in 2013. Both the five museums along the Kunstmeile, established in 2010 (Kunstmeile Hamburg GbR, 2022) and linking several art and culture points of interest in the old city, and the procession of the annual Christopher Street Day Parade contribute to the hotspots in the eastern Altstadt, as revealed by topic analysis. The Rathaus (City Hall) (20) and Jungfernsteig (21), a promenade on the Binnen Alster Lake, are examples of such consistency of presence, and less obviously, so is the Heiligengeistfeld fairground (22) to the west of the Altstadt. By 2019, the thickened arc of hotspots and the southeastern Altstadt landmarks are the dominant features in the heat map, with only the Michel Tower (5) featuring in western Hamburg.

To conclude, by 2014, the equivalent hotspots and the Elbphilharmonie (1) and the Miniatur Wunderland (9) fall within a larger Speicherstadt hotspot as the Elphiharmonie is being constructed on the west end of the Kaiserkai (King’s Quay). The three quays south of the Altstadt Hamburg also have increased in visibility on Flickr, suggesting the “leap over the Elbe” (authors’ translation) and opening development opportunities southward toward Wilhelmsburg, as was the strategic intention of the HafenCity urban development (Hamburg 2005).

Reading the profile of the City of Hamburg using topic analysis and ranking of hashtags of photographs uploaded on Flickr

Seven topics emerge when the three years’ data sets are combined and annotated according to the topic cluster to which each post belongs (Figure 6). The topics are “germany,” “pride,” “juiceclub,” “djjones,” “airplane,” “hafen,” and “travel.” The “hafen” topic is the dominant topic used to tag photographs across all the years within a search conducted on the hashtag “Hamburg.” The respective datapoint map of this topic analysis (Figure 6) shows a logical and expected concentration of photographs related to certain topics.

Figure 6

Topic clusters of unique posts over all three years showing the data points for the “pride,” “hafen,” and “travel” topics. Source: Authors‘ own elaboration, contains information from OpenStreetMap and OpenStreetMap Foundation, which is made available under the Open Database License 2021.

The topic “germany,” due to the fitting method of the NMF, is a rather large repository of photographs with diverse hashtags and topics, so we avoid drawing conclusions from this topic. Of the topics, only “travel” and “airplane” increase; all other topics decrease. The “pride” and “travel” topics are not consistent over the years, with peaks in 2014 and 2019, respectively. Although the “hafen” topic has lost importance in absolute values in 2019, this is not significant in percentage, especially if we consider the “juiceclub” and the “airplane” topics. Our assumption that the “hafen” topic would show increase over the years is not validated by the topic analysis. The “hafen” topic accounts for 32.3%, 30%, and 25% of the tokens (a token is a meaningful element of text), respectively, over the years. It is a heterogeneous constellation in that it is not dominated by limited actors, as, for example, with the “airplane,” “juiceclub,” and “djjones” topics, which, after interrogation, revealed regular posts by particular Flickr users. The spatial visualization of data points (Figure 7) for the “hafen” topic reveals the topic is evenly distributed over the city and not concentrated at the harbor areas only. Further analysis of the “hafen” topic on a year-on-year basis calls for rigorous data cleaning beyond the limitations of this research to draw reasonable and plausible findings.

Figure 7

Distribution of the “hafen” topic in Hamburg. Source: Authors‘ own elaboration, contains information from OpenStreetMap and OpenStreetMap Foundation, which is made available under the Open Database License 2021.

An analysis of the top 30 ranked hashtags and their associated number of photographs in the three years shows the following notable findings (Figure 8). Three historical points of interest appear in the top 30: “altona” (rank position 9), “alster” (rank position 30), and “speicherstadt” (rank position 17). Both “altona” and “alster” have reducing proportions of photographs in the data set; the former reduces from 9.3 to 2 to 1.3 percent of photographs and the latter from 3.7 to 2.1 to 1 percent. However, for “speicherstadt,” the proportions are relatively consistent across the years, at 3.2 to 2.7 to 3.5 percent. The hashtag “elbphilharmonie” has rank position 67 and an increasing proportion of photographs in the data set, from 0.54 to 1.63 to 1.7 percent.

Figure 8

Number of photographs for the top 30 hashtags. Source: Authors’ own elaboration.

Authors of posts make deliberate choices in semantics to convey sentiment and associations. We identified 976 photographs with the hashtag “elbphilharmonie.” This data set and related text proved too small, in this instance, to use natural language processing methods to deliver a reasonable sentiment; nevertheless, the following serves as an illustration of what is possible. A manual review of the comments and data set for the “elphilharmonie” revealed an example for an obviously negative criticism in a caption “faulty design” associated with hashtags as follows: “‘blue, abstract, reflection, building, glass, architecture, square, crane, hamburg, architektur, mirrored, blau, spiegelung, gebäude, glas, hafencity, elbphilharmonie.’” Posts with positive sentiment exist despite the initial controversy surrounding the building during its early construction phases. Eighteen photographs with the caption “another giant in the Hamburg seaport” were posted on the same day in 2014, and several posts over four days in 2014 contained the hashtag “Blaumeise” (English from Dutch: Blue maiden); both of these examples can be interpreted as the authors having applying affectionate metaphors to the building.

Applying deep learning image recognition methodology to the data set of photographs we collected that can identify architectural features of the Elbphilharmonie and thus extract photographs that have not used the hashtag “elbphilharmonie” would serve to increase the size of the data set. Similarly, it would increase the prospects of reasonable results for classic sentiment analysis of texts. Further developing features from the Google API Landmark project for automated labeling of photographs would enable the sentiment analysis of photographs only to be taken up in future research.

To conclude, the use of topic analysis and spatial visualizations iteratively resulted in the differentiation of data between popularity of points of interest and posts by influencers. The “hafen” topic is consistently associated with Hamburg across the years, thus verifying the harbor as a large part of Hamburg’s identity. The association of the hashtag “elbphilharmonie” to the topic clusters “pride” and its subtopic “july” suggests appropriation by particular actors of the exceptional architecture—an example of the unintended impacts of these built artifacts beyond the seeded narratives of city officials and architects who explain the architecture and its intention.

Viral effects of Elbphilharmonie?

Indicators for viral effects on social media were considered to relate to the volume of photographs, the multiplication by sharing and resharing of photographs, the circulation of adapted photographs of the exceptional architecture or an iconic photograph of it, and the photographs that replicate the official photograph. Extracting evidence for these effects requires deep learning image recognition and hashtags—both of which make photographs searchable—and the metadata related to interactions (i.e., likes and comments related to the social media posts). The Flickr platform presented limitations for analysis of such indicators, as it does not have a sharing or reposting function similar to that of Twitter, and deep learning image recognition methods beyond the scope of this research would be required for identifying features in photographs to trace adaptations or replications of an original photograph.

Our analysis revealed that the hashtag “elbphilharmonie” increased in volume as expected over the years (Figure 9) despite the peak of Flickr usage in 2014, as other image-based social media platforms, such as Instagram, became more popular. The location of datapoints mapped (Figure 9) shows increased attention to the Elbphilharmonie as an iconic building starting in its construction phase. From 2009, photographs are taken that include views of the building from the banks of the Elbe and by 2014 from close by the site of the building under construction, then by 2019, from inside the building itself or from its plaza—a public platform at 36 meters above ground level—after it became operational in July 2017. The increased volumes of photographs taken from the specific locations, such as Landungs- and Überseebrücken, the Theater an der Elbe, and boats, are likely attributed to the photographing of the Elbphilharmonie and replicate the official photographs. Deep learning image recognition methods in future research would verify the actual proportion of photographs that do this. Also, when comparing unique posts and all photographs, the 23.3% difference for “elbphilharmonie” versus the 97.5% difference for “juiceclub” suggests that posts hashtagged with “elbphilharmonie” have a larger variety of individual producers of photographs than in the case of “juiceclub.” In the latter case, the institution uploads albums of photographs in a single post, while in the case of “airplane,” a single influencer daily posts a handful of photographs.

Figure 9

Mappings of “elbphilharmonie” data points for 2009, 2014, and 2019. Source: Authors‘ own elaboration, contains information from OpenStreetMap and OpenStreetMap Foundation, which is made available under the Open Database License 2021.

The findings in the previous sections referring to topic clustering, ranking of hashtags, the volumes of photographs taken from locations that aim to replicate the official photograph, and the heterogeneous nature of the posts related to the Elbphilharmonie combined are indicative of the viral effects of the Elbphilharmonie. However, the limitations of the Flickr platform did not yield sufficient data to trace the multiplication patterns to categorically confirm this. Further research through natural language processing and social computing methods could do so.

Conclusion and further research

The heat maps reflect the summation of disconnected choices of individuals. Viewed collectively, these individual choices reflect an aspect of how a city is consumed in the digital era. In this instance, Flickr’s data offers personalized data in both visual and textual formats of a place experienced in the city. Hence, data scraped from the Flickr platform can be used to measure popularity of places in a city and trace the development of popularity over time. However, these places are often stimulated by recurring popular events, such as the Christopher Street Day demonstration for LGBTQI, and exhibitions, such as at Miniatur Wunderland. Exceptional architecture often does not play a role in this process at all. This points to the difference between popular places that are consumed and symbolic places that are revered. Popular hotspots in Flickr are not necessarily symbolic of the image of the city. This also holds true for other social media platforms and relates to how these platforms are used and the motivations driving the sharing of photographs and information. The question then is whether and, if so, how such a summation of individual choices goes on to affect how a city is collectively perceived. This question is increasingly pertinent as more people view cities and places virtually than physically. In addition, the virtual experience will likely precede a physical experience for most tourists and visitors of cities. In that sense the meanings, narratives, and myths that people might associate with cities, places, and buildings might be largely preconditioned by the exposure to collective reflections, which are captured on social media platforms facilitating the development of visual representations of new identities (Vale 1992). This means that meanings, narratives, and myths become uncoupled from the political projects that try to steer these through spatial interventions in the first place.

To uncover whether and, if so, how exceptional buildings, such as the Elbphilharmonie, draw attention and then go on to become city icons, we must interrogate the data further to understand how iconicity delivers more than a photo opportunity for “instant gramification” (FitzGerald 2020). Further research would plausibly explain the topic associations, which reference the semiotics of iconic buildings and landmarks in cities besides the potential held in classical sentiment analysis on sufficiently large data sets. Prima facie evidence indicates, by the location and increasing volume of pictures taken from very particular positions replicating its official photograph, that the Elbphilharmonie has become a city icon. Photographs of it carry its symbolism that demonstrates that “it should be possible (…) to create in architecture a new sensuality, emotionality and poetry in architecture. Hamburg is thus looking for its ‘own way’—away from populist masquerade, meager expedient rationalism and supercooled minimalism” (Hamburg 2005, 30 authors' translation). This could spread the myth of generational relevance, as it abstractly reminisces on Hamburg’s heritage and simultaneously promises a growing city whose future (Hamburg 2005) is, however, not necessarily directly related to Hamburg’s port. Further research tracing the distribution, redistribution, repetition, or adaptation of the myth in social media and at a significantly larger scale facilitated by automated scraping tools would confirm this. It is both the symbolism and popularity of the building that encourage the distribution of photographs and contribute to the iconicity of the building. The association by hashtag of “elbphilharmonie” alongside other historical Hamburg points of interest in a relatively short time frame suggests that it too is part of Hamburg’s identity, more than the other new points of interest in the city have achieved, such as Hadid’s Elbpromenade, HafenCity Hamburg University campus, the Elbbrücken, or the popular Miniatur Wunderland. The indicative viral effects of the Elbphilharmonie due to the heterogenous character of the posts are not proliferated by influencers, as is the case with the “juiceclub” or “airplane” posts, which are regular inputs into Flickr. But the who and what driving the fine-grained public opinions and semiotic development of the “elbphilharmonie” are not made evident in this Flickr analysis. On the other hand, in the methodological process of trial and error, we became aware of tools available for future research to access a broader spectrum of social media platforms and better understand the how and when narratives are spread.

Star architecture is a promising field for further investigation in the mining of Big Data to investigate impact in the field of architecture and urban planning. In many parts of the world, public funds are invested in expensive cultural buildings, based on the promise that they multiply value. They are deliberately designed to be photogenic in response to their reliance on ongoing visual media exposure to build their iconicity beyond the initial marketing hype. This kind of research can test some of the assumptions and expectations pinned to these ambitious projects. Besides the automated tools that can mine across social media platforms, image recognition through computer vision would be useful for categorization of vast numbers of photographs, which in turn contributes to collating bigger data sets for more reasonable sentiment analysis and more like comparisons. Notwithstanding, the challenges and limitations associated with the use of such data include the validity of the source data and its respective biases, interpretation of spatial visualizations without interrogating the underlying data, the normalization of the data to improve its plausibility, and vast amounts of data and computing power and virtual storage space needed to draw reasonable conclusions through automated process.

Acknowledgments

We thank the following students and student assistants for their contribution: Miriam Anschütz conducted the topic analysis of hashtags, Charlotte Hugot produced all GIS-based maps with input from Mathias Heidinger, Hanna Krohberger produced the base map of Hamburg, and Maximilian Wiesholler contributed to the development of the web scraping tool.

Comments
0
comment
No comments here
Why not start the discussion?