Localisation of foreign architecture stylistic prototypes based on multidimensional clustering: Public buildings in Jiangnan, China (1840—1949)

Abstract

During the modern period of China (1840—1949), foreign public buildings in the Jiangnan region (represented by Shanghai, Jiangsu and Zhejiang) exhibited a diverse range of architectural styles. To investigate how the architectural language of these foreign buildings was absorbed and integrated into local building typologies, this study employed a multidimensional clustering method to analyse their formal characteristics, aiming to reveal potential differences in the processes of localisation. Through data coding and decoding, two clustering patterns emerged: (1) cluster A: modern-localised, dominated by Western modern architectural prototypes; (2) cluster B: classical-localised, dominated by Western classical architectural prototypes. Further analysis identified two distinct localisation paths: (1) cluster A by modern-localised follows a deepening localisation path, indicating increasing cultural integration; (2) cluster B by classical-localised follows a weakening localisation path, reflecting a persistence of foreign influence. Situated within the broader historical context of modern Chinese architecture, these findings illustrate the evolving architectural discourse on the interplay between locality and foreignness from 1840 to 1949, offering a new perspective for studying the localisation of foreign architectural forms.

Graphical abstract

Keywords

Formal characteristic / Multidimensional clustering / Modern historical building / Jiangnan region / Localisation

1 Introduction

1.1 Research background

After the Industrial Revolution (1760—1840), European countries spread Western ideas and culture globally, influencing architectural styles across Asia, including modern China (1840—1949). During this era, foreign architecture―primarily Western in origin―was introduced to China under the influence of imported Western forces, as well as Western factory architecture introduced by Chinese foreign affairs factions and the private sector (Pan, 2009). The Jiangnan region, including Shanghai, Jiangsu and Zhejiang, became a key area where this foreign architecture took on diverse forms, reflecting a range of Western architectural prototypes. In Jiangnan, public buildings stood out as leading representatives of foreign architectural expression. During the modern period in China, public buildings in Jiangnan reflected traditions and forms from numerous historical periods and regions―such as Romanesque architecture, Gothic architecture, Renaissance architecture, Baroque architecture, Southeast Asian colonial architecture, North American colonial architecture, Neoclassical architecture, Eclectic architecture, Art Nouveau architecture, Art Deco architecture, Functionalist architecture, Expressionist architecture―as well as regionally adapted styles like Tudor architecture, Neo-Classical Chinese architecture, Spanish architecture and Vernacular Revival architecture (Chen, 2017). Although these architectural forms possessed distinct prototypical characteristics, they were all subject to localisation within the Jiangnan context, collectively presenting a phenomenon of stylistic convergence through similar design strategies. Against the background of the pluralistic intermingling of modern Chinese architecture, this study aimed to quickly identify the characteristic differences of foreign buildings as they underwent localisation within modern Chinese architecture. By focusing on public buildings in the Jiangnan region, this study sought to launch a thematic study on the localisation of the formal characteristics of foreign architecture in the Jiangnan region of modern China.

1.2 Research on the characteristics of architectural form

In art and aesthetics, the term ‘form’ refers to the appearance, structure, and organisation of a work (Tatarkiewicz, 1980). In contrast, the term ‘characteristic’ typically denotes the manner in which human beings perceive the coherence between linguistic entities within a language; it also assists in recognising the manifestation of coherence across different languages (Corbett, 2012).Furthermore, according to Professor Qi (2000), a Chinese architect once distinguished between the terms ‘form’ and ‘morphology’ in architecture. He believed: ‘As an architectural form, it referred to an externalised structure, such as spatial form, which was relatively fixed, unchanging, and static. The dictionary explanation of architectural morphology referred to a situation or momentum, that is, the study of the evolution and development process of form. People studied its variable and dynamic evolutionary process, which was shaped by human intervention and creativity under certain time and environmental conditions.’ Therefore, compared with the dynamic momentum inherent in architectural morphology, architectural form tends to describe the static, externalised, and intrinsic aspects of architectural expression.

1.3 Research on localisation of stylistic prototype

As Xia (1999) suggested, localisation implied the establishment of a root or foundational connection to local culture. In architecture, this concept corresponded to the adaptive transformation of foreign styles to reflect regional identity (Cui, 2007). In this paper, from the perspective of architectural stylistic evolution, the essence of localisation lay in interpreting the stylistic development of Western architecture through a local lens. It not only pertained to the transformation of foreign architectural forms within the Jiangnan region, but more importantly, emphasised the significance of examining this evolution from a local perspective. This approach highlighted how public buildings in Jiangnan during modern China evolved under the dual influence of diverse Western architectural styles and traditional Jiangnan architectural design, thereby resulting in a hybrid stylistic form.

A style, whether architectural or otherwise, establishes boundaries that dictate what is and is not permissible for the person or object embodying it. Departing from these boundaries results in a loss of the style’s distinctiveness, leading to forms or behaviours that become less precise and even sloppy, thereby disqualifying them from being considered part of that style (Edwards, 2013). Therefore, as a disciplinary norm imposed by stylists, architectural style represents a form of rationality to which architecture voluntarily submits.

In recent years, many scholars have discussed localised variants by comparing stylistic prototypes of formal architectural characteristics. For example, Yue and Hu (2024) examined the origins and features of the English Rococo-Chinese style, highlighting its significance as a product shaped by Chinese aesthetic influences. Kadir et al. (2019) challenged the conventional definition by linking it to the aristocratic realm. They argued that, although local architectural forms had undergone technical modifications, they fundamentally retained the essence of the Malay architectural archetype, which embodied attributes of ‘classicism’ and ‘elegance’. Chen (2017) summarised three principal architectural styles that influenced the formation and development of architectural forms in Suzhou, Wuxi and Changzhou from the perspective of industrial and commercial cultural influences during the period of modern China, including classical imitation, early modernism and traditional vernacular improvement styles. Most of the architectural styles were derived from Western influences, such as North American colonial, Southeast Asian colonial, eclectic, Art Nouveau, Decorative, Expressionism, and so on.

Although existing studies have produced substantial insights into buildings with clear stylistic characteristics, this tendency to focus on well-defined stylistic boundaries may have overlooked the ambiguous, transitional forms of architectural styles found in diverse contexts, as well as those ‘boundary buildings’ located at the intersection of multiple styles, which are not only important manifestations of the evolution of architectural culture but also crucial for understanding the dynamic development of form and regional adaptation through localisation. In this context, ‘stylistic characteristics’ referred to design elements that clearly signalled an orientation toward specific Western architectural traditions. Examples included Roman columns, pointed arches or rose windows formed within ribbed vaults, typically arranged according to classical principles of symmetry. Such elements could be empirically associated with particular stylistic traditions―such as Romanesque or Gothic architecture―based on their formal features and compositional logic. Meanwhile, ‘stylistic boundaries’ corresponded to the very definition of style itself. As such, architectural style, particularly as conceptualised and codified by style theorists, imposed a form of disciplinary rationality, which meant buildings that conformed to a specific style were thus perceived as consciously aligning themselves with this rational system and its inherent formal constraints.

During the period of modern China, the localisation of foreign public buildings in the Jiangnan region was manifested in the coexistence of multiple Western architectural style prototypes and the incorporation of design features drawn from traditional Jiangnan architecture. For example, the Anglican Cross Church in Wuxi, Jiangsu Province, retained the understated spatial qualities characteristic of traditional Jiangnan buildings, while simultaneously exhibiting Romanesque facade elements (Chen, 2017) and Gothic-style pointed arches (Liu, 2006). This phenomenon showed that the localisation of foreign public buildings in modern Jiangnan was not a straightforward transplantation of a single Western architectural style prototype, but involved the integration of multiple Western architectural styles, represented as ‘N’, with traditional Jiangnan architectural elements, represented as ‘1’. Through this synthesis, Western architectural styles were recombined and reinterpreted from a local perspective, forming a mixed architectural style by ‘N+1’ with anti-colonial sentiments and multiple architectural style characteristics.

Given the manifestation of this indeterminate architectural style and its formal identity, existing research in China tended to interpret it as a form of eclectic architecture. For example, the St Emil Church, built in Qingdao, Shandong Province in 1934, combined the characteristics of Gothic and Romanesque church architecture. Researchers combined the perspective of architectural stylistics and the analysis of architectural morphology to classify this church as an eclectic style (Zhang, 2006).

However, in the context of Western architectural prototypes, ‘eclectic architecture’ typically referred to the integration of multiple Western architectural forms within the same building type (Bodart, 2003), rather than the incorporation of local design elements. In the modern period of China (1840—1949), by contrast, the term often referred to the fusion of European and American architectural forms with traditional Chinese architecture (Cheng and Neisch, 2023). Some scholars even described the eclectic architecture of modern China as ‘East-meets-West’ architecture, emphasising structures that intricately blended elements of both Chinese and Western design traditions (Guo and Luo, 2015). Nevertheless, the term ‘eclectic’ remained largely a convenient label, and the process behind these ambiguous architectural style forms continued to pose challenges for precise analysis and interpretation.

Therefore, based on the exclusivist approach to architectural style identification, it was difficult to effectively study the localisation of architectural styles that exhibited mixed formal characteristics, such as the eclectic architecture of modern China, often referred to as ‘East-meets-West’ architecture. In response to this limitation, this study advocated the use of an analytical framework based on the formal characteristics of architectural ontology to study the localisation transformation process of foreign buildings that incorporate multiple architectural style prototypes.

2 Literature review

Multidimensional data clustering analysis, as one of the important concepts in data mining, can be applied through linear or non-linear algorithms to scientifically interpret, analyse and utilise large and complex multidimensional datasets (Legendre and Legendre, 2012). This analytical approach has also been applicable in the field of architecture, particularly for identifying the characteristics of architectural forms influenced by various stylistic prototypes. Multidimensional data clustering constitutes a branch of quantitative analysis of multidimensional data and involves two main components: coding by multidimensional data and decoding by cluster analysis.

2.1 Coding by multidimensional data

Multidimensional data serves as a foundational resource, containing rich information and providing a robust database for data analysis (He, 2023). The technical approach to handling multidimensional data can be divided into two main aspects: (1) data description, which refers to exploratory data structure modelling; and (2) data encoding, which involves transforming the data into formats that can be automatically recognised and processed by computers.

Architecture can be likened to a complex living organism. If its formal constituents are traced back to their origins, just as genes are categorised in molecular biology, the differences among coexisting and intertwined architectural forms can be clearly understood. At present, several research approaches in the field of architecture draw inspiration from biology. For example, the Japanese Metabolism school has long pursued this analogy, viewing the city as a living organism and architecture as the gene of the city (Miles, 2013).

The propagation of architecture mirrors the propagation of life, the expression of architectural forms parallels the expression of biological traits, and the fusion of different architectural styles resembles the reproduction of new life by biological hybridisation. This analogy enables us to understand architectural form as a manifestation of a biological system―one that can be deconstructed, analysed quantitatively and traced back to its roots within an architectural tree of life.

At present, many scholars in China and abroad have achieved notable progress in research on the coding of architectural forms. For example, in the study of the formal characteristics of sacred historical architecture, the Italian art critic Gillo Dorfles was among the earliest to adopt a coding-based approach. Dorfles proposed the concept of the Iconic Image Code through the repetition of a series of fixed forms in sacred historical architecture, emphasising the importance of critically interpreting these codes (Dreyer, 2010). Similarly, Eisenman (2004), a prominent figure in modernist deconstructivism, applied coding theory to the biological metaphor of architectural concepts, arguing that ‘Very often, when A-B-A or A-B-B-B-A relations are used to represent proportionality in plans and sections, these symbols do not recognise other relations, such as positional relationships.’ Therefore, according to Eisenman, complex architectural conditions such as thickness, height, material and surface could be reduced to simplified geometric lines in formal analysis. These, in turn, could be interpreted as metaphors for simplified architectural forms and their morphogenetic meanings through topological and fractal theories. This perspective contributed significantly to advancing the application of coding theory in architectural form analysis. The relevance of coding in architectural style analysis has also drew the attention of Dr Andrea Gleiniger. Gleiniger and Vrachliotis (2012) argued that architecture, as an aesthetic tool and ideological orientation, has broad public value and is applicable to most people, while specific coding expresses different political, social and artistic attitudes in architectural design. In recent years, many scholars in China and abroad have continued to develop research on the coding of architectural design, including studies on the coding of architectural shape and contour (Wei et al., 2021), spatial structure (Dursun, 2012) and facade elements (EL-Mahdy, 2025), among others.

Therefore, the findings of many scholars demonstrated the feasibility of encoding architectural form characteristics. The complex formal characteristics of public buildings among foreign buildings in the Jiangnan region during the modern Chinese period covered, to varying degrees, the genetic expression of prototypes, which further influenced the typological structure of localised buildings through stylistic blending.

As such, applying these well-established coding methods to the study of the localisation of foreign public buildings in modern Jiangnan allowed for a more objective understanding of the design evolution. By deconstructing and encoding architectural forms through machine learning, it became possible to clarify how various architectural style prototypes shaped the development of foreign public buildings in the region during this period.

2.2 Decoding by cluster analysis

Cluster analysis refers to the grouping of data objects based solely on the information contained within the data that describes the objects and their relationships. The goal is to ensure that objects within the same group are similar (related) to each other, while objects in different groups are dissimilar (unrelated) (Tan et al., 2006). Cluster analysis organises data into meaningful or useful groups (clusters), such as separated clusters, centre-based clusters, proximity-based clusters, density-based clusters and conceptual clusters. According to the cluster type set results, the clustering results can also involve hierarchical clustering, partition clustering, exclusive, overlapping, non-exclusive, fuzzy clustering, complete clustering and partial clustering (Tan et al., 2006). Cluster analysis of multidimensional data is typically used to deal with datasets containing multiple characteristics, and may also involve the analysis of multiple cluster types or clustering results to identify underlying patterns or structures of the object of study in a high-dimensional space.

According to Turchin (2018), a professor at the University of Connecticut in the United States, real societies are always made up of many different qualitative and quantitative factors that interact with each other in a very complex way and therefore require the intervention of mathematical theories. Cluster analysis, as a method of statistical analysis, is capable of concentrating and refining cluttered information and identifying patterns inherent in the chosen object of study. In recent years, there have been increasing research results on the use of cluster analysis methods to analyse the characteristics of architectural forms, including studies of the similarity of building shapes (Xia et al., 2020), research on rapid identification of architectural elements (Shalunts et al., 2011), and so on. However, there have been fuzzy forms of architectural localised formal features under the influence of multiple architectural prototypes, and the application of cluster analysis in assisting in decoding the localised laws in architectural multidimensional data information has still been relatively limited.

Therefore, this study examined the public buildings among the architectural forms in the Jiangnan region during modern China. Based on the logic of coding and decoding, it developed a multidimensional approach to coding and decoding through cluster analysis of the formal features of the architectural samples. The aim was to facilitate computer-assisted exploration of the localisation patterns of the ambiguous forms of the foreign buildings in Jiangnan during the modern period, shaped by various architectural prototypes and cultural influences.

3 Methodology

This methodology drew inspiration from the conceptual logic of molecular biology, treating architecture as a living organism. It systematically disassembled and encoded architectural stylistic design in multiple dimensions, ultimately aggregating and clustering the data for comprehensive analysis. The strength of this approach lay in its ability to reinterpret architectural styles purely through design principles. Furthermore, it offered a possible quantitative method for addressing architectural design challenges shaped by the interaction of multiple stylistic prototypes. The multidimensional data clustering methodology followed a systematic process comprising six key steps:

Step 1: Data collection. Historical data and architectural drawings of selected public buildings in the Jiangnan region of modern China were gathered through field surveys.

Step 2: Data coding. Based on the collected historical data and architectural drawings, a coding dimension was established, including four key dimensions―plan modelling, facade modelling, material structure and decorative pattern―along with eight specific variables. Two coding methods were adopted: set-style coding for abstract formal concepts and label-style coding for concrete formal elements.

Step 3: Data decoding. A suitable algorithm, such as the Longest Common Subsequence (LCS) algorithm, was selected for data decoding to identify similarities among the architectural samples.

Step 4: Data visualisation. To represent the clustering results, appropriate visualisation techniques, such as dendrograms, were employed, which enabled a clear display of the relationships among the architectural samples.

Step 5: Identification of style prototypes. Based on the results of the multidimensional data clustering analysis and through comparison with documented Western and Chinese architectural style prototypes, the architectural style prototypes corresponding to each cluster of modern Jiangnan public buildings were identified.

Step 6: Interpretation of stylistic prototypes localisation. Based on the identified architectural sample stylistic prototypes, a reasonable interpretation was made of the localisation differences, path and reasons for the foreign architectural style of public buildings in the Jiangnan region of modern China (1840—1949).

By applying the multidimensional clustering method, it became possible to conduct quantitative research for effectively tracing the localisation path of foreign architectural styles in the modern Jiangnan region, and also to digitise and store the relevant architectural sample data. On the one hand, this move contributed to the construction of a regional architectural digital database; on the other hand, it provided support for quantitative methods in research aimed at analysing the design trends of a large number of architectural samples (Fig. 1).

3.1 Data collection

3.1.1 Collection area

In this study, the Jiangnan region, represented by Shanghai, Jiangsu and Zhejiang, was selected as the data collection area for the following reasons. Firstly, during the modern period of China, foreign colonisers coveted the region’s superior geographical conditions and signed treaties with local governments to open these cities to foreign access at an early stage. As a result, the Jiangnan region became a primary hub for foreign cultural influence in China (Table 1). In these colonial areas, a large number of foreign public buildings featuring prototypical Western architectural styles were constructed, including religious buildings, cultural and educational buildings, commercial and trade buildings, medical buildings, office buildings and transport buildings. Secondly, due to its developed economy and rich cultural heritage, the Jiangnan region provided excellent conditions for the preservation of historical buildings. This advantage ensured the availability of sufficient and reliable data, thereby supporting the research efforts (Fig. 2).

3.1.2 Collection criteria

The collection criteria for building samples included three aspects. Firstly, the collection range of architectural samples referred to cultural heritage buildings that had been listed as national, provincial and municipal cultural protection units in Shanghai, Zhenjiang, Suzhou, Hangzhou, Ningbo, Nanjing, Wenzhou, Lianyungang (formerly known as Haizhou), Nantong (formerly known as Tongzhou) and Wuxi in recent years. Secondly, certain heritage buildings were excluded due to practical constraints such as deterioration from age and wartime damage, poor maintenance, incomplete architectural drawings and the inability to inspect their current condition. Thirdly, the building samples collected in this study were all locally constructed public buildings influenced by foreign cultures. This influence was reflected in the following two aspects. On the one hand, the building samples were built by foreign architects, such as the Dongjiadu Catholic Church in Shanghai and the former site of Zhejiang Customs in Ningbo, both of which were built by foreign architects in the modern period of China. On the other hand, the architectural samples were designed by local architects who had been influenced by foreign culture, typically those who had studied abroad and returned to China, such as the Xiaguan Railway Station Reconstruction Project and the Former Central Stadium.

3.1.3 Collection result

The collection result included ninety-one foreign public building samples that met the criteria, comprising six categories: religious building, office building, trade building, medical building, cultural and educational building and transport building (See Appendix A). The ninety-one foreign public building samples in this study followed strict spatial, temporal and typological boundary parameters. Spatially, the study focused on the Jiangnan region (Shanghai, Jiangsu and Zhejiang), as this area represented the most concentrated manifestation of foreign architectural influence in modern China. Temporally, the research scope was limited to 1840 to 1949, corresponding to the period from the First Opium War to the founding of the People’s Republic of China, during which the localisation of Western architecture in China was most active. Typologically, only public buildings (religious, cultural-educational, trade, medical, office and transport) influenced by foreign cultures were included, as they reflected direct cultural exchanges with the West. Furthermore, only buildings that (1) had wellpreserved physical forms, (2) possessed complete architectural drawings and (3) were listed as protected cultural heritage at national, provincial or municipal levels were selected. These parameters defined the boundary of this study and ensured comparability among the selected samples (Fig. 3).

3.2 Data coding

3.2.1 Coding dimension

The selection of the eight specific variables in this study was based on the hierarchical logic of architectural design. Firstly, since the building samples were historical structures, it was essential to reference architectural historical data and drawings. Architectural drawings served as the primary medium for transforming architectural concepts into visual forms, and they also functioned as intuitive information carriers that enabled architectural design proposals to be understood and reviewed by others (Ma et al., 2023). Compared with other forms of documentation, architectural drawings were among the most accessible and reliable sources of architectural information for researchers and stakeholders, and they were also the information carriers that could record and convey the original characteristics and appearance of a building in a relatively truthful, complete and objective way. Furthermore, the fundamental principles of architectural design―solidity, functionality and aesthetics―were translated in this study into three corresponding architectural elements: loadbearing, enclosure and decoration. Accordingly, four key dimensions in architectural historical data and drawings were selected for systematic analysis: plan modelling, facade modelling, material structure and decorative pattern (Fig. 4).

Under the analysis of plan modelling: (a) coding of the plan shape contour, (b) coding of the plan spatial form;

Under the analysis of facade modelling: (c) coding of the facade shape contour, (d) coding of the facade detailed element;

Under the analysis of material structure: (e) coding of the building’s main material structure, (f) coding of the building roof material structure, (g) coding of the building door and window material structure;

Under the analysis of decorative pattern: (h) coding of the building decorative pattern theme.

3.2.2 Coding method

To facilitate machine identification and cluster analysis of the building samples, this study used two coding approaches: set-style and label-style, which corresponded to the eight specific variables of the data recorded in the multidimensional dataset. While the eight specific variables were quantitatively encoded for clustering purposes, they were originally derived from qualitative aspects of architectural form. The study adopted a form-based transformation strategy that converted qualitative characteristics into quantifiable variables in architecture. Among them, two categories were adopted, corresponding to two coding methods. The first category includes abstract formal concepts, such as plan shape contour, plan spatial form and facade shape contour, which were quantified through formal analysis of architectural drawings and encoded using set-style quantification. The second category included concrete formal elements, such as facade detailed elements, main material structure, roof material, door and window materials and decorative pattern themes, which were visually explicit and readily identifiable through direct observation, and were encoded using label-style quantification. By integrating these two coding approaches, the study effectively balanced quantitative structure with qualitative insight, enabling statistical analysis of formal characteristics in the classification of foreign public buildings in the Jiangnan region during modern times of China.

(1)　Set-style

Set-style coding focused on the quantitative transformation of architectural form into structured numerical data. Specifically, set-style coding was presented as a sequence of collections of numbers. Set-style was suitable for the data records of the following three-dimensional characteristics: (a) coding of the plan shape contour, (b) coding of the plan spatial form, (c) coding of the facade shape contour. Set-style data coding was suitable for the data records of the following three-dimensional characteristics: (a) coding of the plan shape contour, (b) coding of the plan spatial form, (c) coding of the facade shape contour.

For the data coding of the building shape contour, to improve the type accuracy of the data results, this study used a coding method based on the number of turns and the angle of turns, which was applied to ‘a. Coding of the plan shape contour’ and ‘c. Coding of the facade shape contour’. In this case, ‘turn’ refers to the tendency of discontinuous straight lines between ordered nodes in the shape contour of a building, and ‘angle’ refers to the inward angle of the graph formed by a turn at any point in the shape contour of a building (internal angle). Since buildings are characterised by multiple turns, their shape contours are enclosed by multiple angles. Therefore, by specifying the number of turns and the angle of turns in the shape contour of a building, the basic characteristics of the shape contour of a modern Jiangnan foreign building can be effectively revealed. The specific operation of graphic coding was as follows: (1) judge the number of turning angles of architectural shapes―this step determined the number of sequences of the point set {P_m-1, P_m, P_m+1}; (2) determine the internal turning angle of the building figure and correspond it to the code of its turning angle, where each 45° turning angle of the internal angle was a unit and the code was recorded (Table 2).

To unify the coding format, this study processed the building samples into regular graphics and began from the internal turning angle at the upper left corner of the graphic outline, then carried out the sequence coding of the building plan shape contour data according to the turning point and the turning angle of the internal angle. For example, in the simple building form in Fig. 5, the square plan and facade were both formed by four turns with internal angles of 90°, so the point set was represented as {2, 2, 2, 2}, starting from the upper left corner of the figure. In contrast, the complex building form had a polygonal plan and facade with twelve and eight turns respectively, and the internal angles of 90° and 270° were formed by different orders of combination. Thus, the plan point set was represented as {2, 2, 6, 6, 2, 2, 2, 6, 2, 2, 6, 2}, starting from the upper or upper left corner of the figure, and the facade point set was represented as {2, 2, 6, 2, 2, 2, 2, 6}, starting from the upper left corner of the figure (Fig. 5).

For the data coding of the building plan spatial form, this study used a coding standard that recorded the spatial connectivity of internal space units to represent the layout of building spaces. In this case, the analysis was based on the building plan. In the theory of spatial syntax, the basic analytical variables of spatial structural characteristics are connectivity, control, depth and integration (Hu and Liu, 2020), where connectivity refers to the number of nodes neighbouring a certain node, which is the connectivity of the node, and also represents the permeability of the space (He, 2017). The specific quantification rules were as follows: if a space was connected to one space, its connectivity was one; if connected to two, the connectivity was two. Therefore, each number represented the connectivity of each space with the adjacent space. In the spatial form coding method, the arrangement was generally in descending order. Figure 6 shows the data coding process of the building plan spatial form.

The following presents an example of set-style data coding. According to the coding rules, the building plan shape contour of the Former Old Post Office in Zhenjiang was coded as {2, 2, 2, 2}; the building facade shape contour of Gospel Church in Zhenjiang was coded as {3, 3, 2, 2, 3}; and the building plan spatial form of the Former Xijinqian Silk Industry Association in Wuxi was coded as {3, 1, 1} (Fig. 7).

(2)　Label-style

Label-style coding focused on the visual identification and manual classification of stylistic features of the building samples. In terms of typological application, label-style was suitable for the data records of the following five-dimensional characteristics: (d) coding of the facade detailed element, (e) coding of the building main material structure, (f) coding of the building roof material structure, (g) coding of the building door and window material structure and (h) coding of the building decorative pattern theme.

For the data coding of building facade detailed elements, the detailed elements in modern Jiangnan foreign buildings were mainly shown as arch (A), column (C), beam (B) and wall (W) (Fig. 8).

For the data coding of the building main material structure, the overall material structure in modern Jiangnan foreign architecture roughly went through three stages: brick-wood hybrid structure, brick-reinforced concrete hybrid structure and steel and reinforced concrete frame structure. These were the three stages in the transformation of material technology that provided a variety of styling possibilities for modern architecture in China (Lai and Tong, 2017). According to the classification criteria of brick-wood hybrid structure (B, W), brick-reinforced concrete hybrid structure (B, S, C) and steel and reinforced concrete frame structure (S, C), label-style data coding could be carried out (Fig. 9).

For the data coding of the building roof material structure, the roof materials in modern Jiangnan foreign buildings were mainly manifested in four common types: handmade tile (T), machine-made tile (M), modern panel (P) and poured cement (C). The roof framing structures mainly include the lifted-beam type (L), column-and-tie-beam type (C), lifted-beam-penetrating-double combined type (L, C), serrated-truss type (S), triangular truss type (T), arc truss type (A), frame structure (F) and row-rack structure (R) (Fig. 10).

For the data coding of the building door and window material structure, the door and window materials in modern Jiangnan foreign buildings were mainly wooden glass doors and windows (W) and metal glass doors and windows (M). In contrast, the door and window structures mainly included flat-arched transoms (F), curved-arched transoms (A), pointed-arched transoms (P) and non-transomed transoms (N) (Fig. 11).

For the data coding of the building decorative pattern theme, the decorative patterns in modern Jiangnan foreign buildings, according to the initial letter code of the pattern theme category, mainly included the following nine categories: abstract line pattern (G), lens pattern (L), plant pattern (P), craftwork pattern (C), animals pattern (A), artefacts and magic treasures pattern (M), human story pattern (H) and word pattern (W). In addition, there were also some designs without patterns, called ‘none pattern’ (N) (Fig. 12).

The following presents an example of label-style data coding. According to the coding rules, the building facade detailed elements of the Former Post Office in Ningbo were coded as A, C; the main material structure of the Former County Library in Wuxi was coded as B, W; the roof material structure of St Francis Xavier Church in Shanghai was coded as T, T; the door and window material structure of the Former Christ’s Hospital in Zhenjiang was coded as W, F, A; and the decorative pattern themes of Yangjiaqiao Catholic Church in Suzhou were coded as G, P, M, H, W (Fig. 13).

3.3 Data decoding

Clustering algorithms are a class of algorithms that divide data into groups (i.e., clusters) so that data points within the same cluster exhibit high similarity and data points between different clusters have low similarity. Common clustering algorithms include K-Means, Hierarchical Clustering, DBSCAN (Density-Based Spatial Clustering of Applications with Noise), OPTICS (Ordering Points To Identify the Clusters), DBSCAN (Density-Based Spatial Clustering of Applications with Noise) and Mean Shift. These algorithms vary in applicability depending on the specific characteristics of the dataset and the intended analytical goals. To avoid the loss of data accuracy in high-dimensional spaces, which often results from dimensionality reduction, this study selected a clustering algorithm compatible with both abstract and concrete encoded data. Specifically, it employed a dynamic programming algorithm based on the Longest Common Subsequence (LCS) to process the results of the two types of data encoding.

The Longest Common Subsequence (LCS), which is widely used in speech processing and text data comparison, measures the degree of similarity between two strings by the length of their longest common subsequence, which is the basic idea of the method (Li et al., 2015). Generally speaking, the longest common subsequence algorithm is capable of comparing the similarity between two sequences while keeping the length and order of the two sequences unchanged. Dynamic programming algorithms are often used to solve the longest common subsequence problem for two sequences. One general method for calculating the longest common subsequence is as follows: taking two sequences, x and y, as an example, with a two-dimensional array f(i,j) denoting the length of the longest common subsequence before the i element of x and the j element of y, the algorithm is as follows:

\begin{array}{}& f(i\mathrm{,}j)=\max \{\begin{array}{l}f(i-\mathrm{1,}j-1)+\delta (x_i\mathrm{,}{y}_j)\mathrm{,}\\ f(i-\mathrm{1,}j)\mathrm{,}\\ f(i\mathrm{,}j-1)\mathrm{,}\end{array}\end{array}

where

\begin{array}{}& \delta (x_i\mathrm{,}{y}_j)=\{\begin{array}{c}\mathrm{1,}{x}_i=y_j\mathrm{,}\\ \mathrm{0,}{x}_i\ne y_j\mathrm{.}\end{array}\end{array}

At this point, the largest number in the array represents the length of the Longest Common Subsequence (LCS) of x and y (Qu et al., 2011). The LCS algorithm was suitable for this study’s application to both set-style data and label-style data because it satisfied two essential conditions: first, that the two sequences exhibited an optimal substructure; and second, that they contained overlapping subproblems.

Specifically, the LCS algorithm was capable of identifying common subsequence within different datasets. For both set-style and label-style data, it could detect shared elements between different sequences. For example, in set-style data, if two sequences were {1, 2, 1} and {1, 1}, the LCS identified {1, 1} as the longest common subsequence. Similarly, in label-style data, such as sequences W, A and W, F, A, the LCS yielded W, A as the common part. This enabled the algorithm to measure and compare the degree of similarity between different architectural feature sets.

Secondly, the LCS algorithm was adaptable when dealing with sequential data; it could retain the sequential information within the data and identify the same number or order of numbers, rather than simply comparing the sequences in isolation. This was also very important when dealing with set-style data and label-style data.

Therefore, focusing on the formal characteristics of modern Jiangnan foreign public buildings, this study recorded the set-style data in three aspects―(a) coding of the plan shape contour, (b) coding of the plan spatial form and (c) coding of the facade shape contour―as well as label-style data in five aspects―(d) coding of the facade detailed element, (e) coding of the building’s main material structure, (f) coding of the building roof material structure, (g) coding of the building door and window material structure and (h) coding of the building decorative pattern theme―for analysis by the dynamic programming algorithm (See Appendix A). The data were analysed by a dynamic programming algorithm that utilised the longest common subsequence.

The clustering process was as follows. Firstly, through the two encoding methods mentioned in ‘3.2.2 Coding method’, the original data of eight specific variables for ninety-one public buildings under four key dimensions (plan modelling, facade modelling, material structure and decorative pattern) were recorded respectively. Among these, the three aspects of (a) coding of the plan shape contour, (b) coding of the plan spatial form and (c) coding of the facade shape contour were involved, and the five aspects of (d) coding of the facade detailed element, (e) coding of the building’s main material structure, (f) coding of the building roof material structure, (g) coding of the building door and window material structure and (h) coding of the building decorative pattern theme were marked. Secondly, the similarity matrix of the original data of eight specific variables under four key dimensions of ninety-one public buildings was calculated respectively through the LCS algorithm. This included the similarity matrix for plan shape contour, plan spatial form, facade shape contour, facade detailed element, building’s main material structure, building roof material structure, building door and window material structure and building decorative pattern theme. Eight similarity matrix tables were then derived from the original data of the eight specific variables. Furthermore, for the convenience of statistics, the influence weights of the eight specific variables under the four key dimensions of public buildings were assigned in a 1:1 ratio, and the final similarity matrix was then calculated (using the average value) to obtain the final similarity matrix of ninety-one public buildings. Finally, the dendrogram clustering results were obtained based on the final similarity matrix.

4 Findings

A dendrogram is a hierarchical view that enables data visualisation and facilitates the comparison of different levels of classification. Dendrograms use colour and proximity to represent categories, making them more effective than many other chart types for displaying large datasets. In a dendrogram, the height of the vertical lines represents the distance or dissimilarity between the data points being linked: the closer the distance, the lower the vertical line and the higher the similarity between clusters; the further the distance, the higher the vertical line and the lower the similarity between clusters. The following figure illustrated the clustering results based on a comprehensive analysis of the similarity matrix for the localised form characteristics of foreign public buildings in the modern Jiangnan region. Based on the clustering results in the dendrogram, it could be seen that there were two distinctly different types of clusters in the expression of the formal characteristics of the localisation of public buildings in the modern Jiangnan region: cluster A and cluster B. Building on the cluster results of formal characteristics of foreign buildings in the modern Jiangnan region, derived through a multidimensional data clustering method, this section further examines the specific manifestations of these characteristics in cluster A, cluster B and several sub-clusters. The analysis is structured around four key dimensions: plan modelling, facade modelling, material structure and decorative patterns, using eight specific variables as guiding elements. The findings provide valuable insights for swiftly identifying differences in design performance among the clustered building samples (Fig. 14) (See Appendix A).

Building on the descriptions of architectural prototypes found in existing books, archives and academic articles, and integrating insights from previous studies, this research further examines the prototypical tendencies in the formal characteristics of the building samples in cluster A and cluster B. Following the clustering results, it is necessary to consult and compare these with established criteria from existing research for defining the characteristics of architectural style prototypes. In recent years, many scholars have explored the prototypical characteristics of modern architecture in the Jiangnan region, particularly focusing on the influence and expression of Western architectural prototypes. Accordingly, based on the clustering results of the formal characteristics of public buildings in modern foreign architecture in the Jiangnan region, cluster A is found to encompass the formal characteristics associated with five architectural prototypes, namely: (1) Renaissance architecture, (2) Neo-Classical Chinese architecture, (3) Art Deco architecture, (4) Functionalist architecture and (5) Expressionist architecture (Fig. 15).

Cluster B encompasses the formal characteristic expressions of thirteen architectural prototypes, namely: (1) Romanesque architecture, (2) Gothic architecture, (3) Renaissance architecture, (4) Baroque architecture, (5) Southeast Asian colonial architecture, (6) North American colonial architecture, (7) Neoclassical architecture, (8) Eclectic architecture, (9) Eurasian architecture, (10) Vernacular revival architecture, (11) Neo-Classical Chinese architecture, (12) Art Deco architecture and (13) Tudor architecture (Fig. 16).

5 Discussion

To better understand of the design logic shown by the clustering results, the discussion examines the localised design differences and their possible causes, as seen in cluster A and cluster B. It will also reflect on how the formal translation of foreign public buildings in modern Jiangnan relates to the process of cultural reconstruction and localisation.

5.1 Cluster A: modern-localised

The architectural samples in cluster A predominantly exhibit formal characteristics derived from Western modern architectural prototypes, such as Art Deco, Functionalist and Expressionist styles, along with some foreign buildings influenced by Renaissance architecture and Neo-Classical Chinese prototypes, reflecting a hybridised architectural identity. Accordingly, cluster A is designated as modern-localised, referring to the negotiation between imported Western modernist aesthetics and local cultural expression.

5.1.1 Differences in cluster A

Despite being composite in origin, the samples in cluster A exhibit a high degree of consistency in material structure and plan modelling, while displaying variations in decorative patterns and facade modelling, indicating a form of selective localisation, wherein imported stylistic systems are internalised through specific formal devices rather than wholesale adoption. Such results, drawn from multidimensional clustering, provide an empirical basis for identifying patterns of formal convergence across diverse stylistic origins―insights often difficult to generalise in purely qualitative typological readings.

Within cluster A, two subgroups (cluster A-1 and cluster A-2) emerge, each reflecting different levels of localisation based on the time of building construction:

The first subgroup, cluster A-1, demonstrates ‘decoration localisation’, wherein Western modern architectural prototype buildings adopt traditional Chinese decorative pattern themes. For example, the Former Auditorium of Central University in Nanjing in 1931 and the Central Hotel in Nanjing in 1927 (Renaissance prototype), as well as the Central Hospital in Nanjing in 1933 and Soochow Hospital in Suzhou in 1922 (Neo-Classical Chinese prototype), integrated Chinese decorative pattern themes such as Chinese traditional plant patterns and the traditional Ruyi cloud pattern. This not only reflects a surface-level hybridisation, but also represents a symbolic negotiation of cultural identity, in which Western architectural language is repurposed to accommodate local aesthetics and values. Such a design strategy of ‘decoration localisation’ endows foreign architecture with local legitimacy and visual affinity by introducing and applying traditional Chinese decorative elements, which is also in line with the trend of cultural hybridity and identity construction in the context of postcolonial aesthetics.

The second subgroup, cluster A-2, advances further into formal localisation, combining both decorative pattern and facade modelling. For example, the facade of the Bank of China Nanjing Branch Building in 1936, which was financed by Kung Hsiang-hsi, Sung Tzu-wen and the Nanjing National Government (Neo-Classical Chinese architecture), adopted a stepped-rectangle facade shape contour with an abstract image of the local horse-head wall steps in the Jiangnan region. Unlike traditional Chinese horse-head walls, which typically appeared on gable walls, those on the Bank of China Nanjing Branch Building were positioned on the main facade, which might have been influenced by the stepped facades of Art Deco high-rise buildings. By 1937, Western modernist architectural thinking permeated the Jiangnan region, leading to architectural forms―especially main facades―that exhibited pronounced geometric characteristics (Dai, 2018). As such, influenced by both a reverence for Western modern architectural thought and a desire to preserve local cultural identity, this hybrid facade style, which combined Chinese and Western elements, became a common feature in the main facade design of modern buildings in the Jiangnan region. Similarly, the Bank of China Building in Shanghai (Neo-Classical Chinese architecture), jointly designed and constructed in 1937 by Lu Qianshou and the British firm Kung Wo, featured a triangular-rectangular facade shape contour with a simplified Chinese four-square spire. Although the building as a whole followed the Western modern high-rise typology, its integration of local architectural pattern themes reflected the same hybrid design philosophy (Fig. 17).

5.1.2 Deepening of localisation path

The difference in the degree of localisation between cluster A-1 and cluster A-2 essentially reflected the varying localisation strategies that were adopted in response to the design demands of distinct historical periods in modern China. This divergence in strategic choices, which shaped different approaches to architectural expression, in turn established a path through which the localisation of foreign public buildings in cluster A―representing a modern-localised process―continued to deepen throughout the 1930s.

The early 1930s in China marked a pivotal period of intersection and transformation between Western classical architectural culture and modern architectural trends (Rowe and Kuan, 2002). At this historical juncture, decorative localisation, which was a design strategy operating at the level of Mannerism, was frequently employed as a transitional device. It had not yet fully departed from the foreign image of Western classical architecture. The localisation approach exemplified by cluster A-1, which reflected decoration-based localisation through the partial incorporation of Chinese decorative patterns within a framework still dominated by Western classical forms, not only served as a symbolic act of cultural embedding―integrating indigenous elements into visual detail to express identity and cultural affinity―but also provided an initial guide and practical foundation for the modernisation of architectural style in China during the late 1930s.

As Professor Zheng (2020) noted: ‘The architectural types of the New Renaissance encompass residences, office buildings, hospitals, museums and more. Their principal feature lies in their use of Italian and French Renaissance architecture as prototypes, drawing upon the forms and decorative elements characteristic of Mannerism.’ Indeed, the Renaissance Revival style enjoyed widespread popularity in countries such as Italy, Germany, France and Britain, symbolising a broader cultural nostalgia for classical history and heritage. In the United States, particularly following the Civil War (1861—1865), this style began to absorb elements from other historic architectural languages, becoming increasingly prevalent after the Philadelphia World’s Fair in 1876 (Gura, 2015). It may therefore be inferred that Chinese buildings from the early 1930s, which exhibited Renaissance or Chinese neoclassical characteristics and included examples such as the Nanjing Central Hotel, completed in 1927, and the National Central University Auditorium, completed in 1931, were likely influenced by this Renaissance tradition. Simultaneously, these structures embedded Chinese decorative patterns or culturally symbolic details in a Mannerist way, reflecting an affirmation of China’s inherent cultural identity and an emotive response to traditional Chinese architectural forms. This design inclination functioned not merely as an aesthetic strategy but as a considered cultural statement.

On 10th January 1935, ten professors, including Xinmian Wang and Bingsong He, jointly published the Declaration on China’s Original Cultural Construction in Volume 1, Issue 4 of the Cultural Construction Monthly. In the Manifesto, there was a heated debate between the local culture school and the westernisation school on the issue of China-oriented cultural construction, which involved consideration of Chinese and Western cultures and the choice of the way forward for Chinese culture (Zheng, 2004). As a result, local ideas became increasingly sophisticated, which may also have influenced the field of architectural design. Consequently, the foreign buildings in cluster A-2, which exemplify the process of form localisation, reflect a more profound level of localised design consciousness, which is evident not only in the treatment of decorative patterns but also in the articulation of the building facades. This localisation design strategy further demonstrates that, by the late 1930s, Chinese architects had essentially developed a dominant position regarding the integration of Chinese and Western architectural design, which signifies a maturing of cross-cultural architectural expression.

Luke and Wu (1936) wrote that ‘first, a successful work cannot disregard practical needs; second, it cannot disregard the contexts of its times; third, it cannot abandon the principles of art; and fourth, it cannot overlook the spirit of culture’. These four propositions demonstrated Luke’s perception of the Chinese architectural spirit during the enlightenment of modern Chinese society. Since Classical Chinese Revival Architecture was a form of Art Deco that reflected an international transition under the influence of Western modern architectural thinking, many modernised Chinese buildings were also regarded as variants of Art Deco. Therefore, some examples of Neo-Classical Chinese architecture, such as the Bank of China Building in Shanghai, built in 1937, and the Bank of China Nanjing Branch Building in Nanjing, built in 1936, not only showed localised decorative patterns, which expressed cultural identity, but also conformed to broader international trends in architectural form, simultaneously highlighting distinctive local design characteristics. These characteristics were evident in the abstract and simplified shapes of the facade modelling, which incorporated traditional elements such as the Chinese four-square and the ‘ma-tou qiang’ (horse-head wall), typical of the Jiangnan region.

5.2 Cluster B: classical-localised

The architectural samples in cluster B predominantly exhibit formal characteristics derived from Western classical architectural prototypes, such as Romanesque, Gothic, Renaissance, Tudor, Baroque, Southeast Asian colonial, North American colonial, Neoclassical, Eurasian and Vernacular revival architectural styles. Accordingly, cluster B is designated as classical-localised, referring to the synthesis between Western classical aesthetics and local cultural expression.

5.2.1 Differences in cluster B

Despite being composite in origin, the samples in cluster B show variation in facade modelling, material structure and decorative patterns―an instance of differentiated localisation derived from multidimensional clustering analysis. This offers an empirical basis for understanding how imported stylistic elements are selectively adapted and reinterpreted within local architectural contexts.

Within cluster B, three subgroups (cluster B-1, cluster B-2 and cluster B-3) emerge, each reflecting different levels of localisation based on the time of building construction:

The first subgroup, cluster B-3, presents a diversified form of localisation, in which Western classical architectural prototype buildings adopt traditional Chinese facade modelling, material structure and decorative patterns. For example, the Yangjiaqiao Catholic Church in Suzhou, built in 1892, and the Xinlan Book Club in Wenzhou, built in 1910 (Vernacular revival architecture), both focused on the localised design of the shape contour in building facade modelling, such as the facade shape contour of local horse-head walls in the Jiangnan region, and the trapezoidal-rectangle facade shape contour combined with a traditional Chinese four-slope hermetic roof. Meanwhile, the material structure of the main bodies of both buildings adopted the traditional Chinese hybrid construction of brick (or stone) and timber. The roofs were covered with traditional handmade Chinese tiles, and the windows were constructed using traditional Chinese wooden frames. In addition, the localised design of the decorative patterns used a greater number of local thematic decorative patterns, which included traditional Chinese word patterns, artefacts and magic treasures patterns, plant patterns, and others.

The second subgroup, cluster B-2, however, exhibits a weakened form of localisation, focusing on the localised design of the material structure and decorative pattern. For example, the Former County Library in Wuxi, built in 1911 (Southeast Asian colonial architecture), and the Sicheng Hall in Hangzhou, built in 1927 (Eurasian architecture), both focused on the localised design of the building material structure. The main bodies of the buildings adopted a traditional Chinese hybrid structure of brick (or stone) and timber, the roofs were covered with traditional handmade Chinese tiles and the windows were made of traditional Chinese wooden frames. Meanwhile, the localised design of the building’s decorative patterns also used local thematic decorative patterns, which included traditional Chinese plant patterns, word patterns and other culturally symbolic elements.

However, the third subgroup, cluster B-1, presents the weakest form of localisation, which focuses only on the localised design of the building facade modelling, often reflected in the building facade shape contour. For example, the trapezoidal-rectangle facade shape contour of the former site of the Former National Resources Commission of the Nationalist Government in Nanjing (Vernacular revival architecture), built in 1947, exhibited the characteristics of a traditional Chinese four-slope hermetic roof. Meanwhile, the trapezoidal-rectangle facade shape contour of the Suzhou University Gym (Romanesque architecture), built in 1937, also featured a gentle traditional Chinese four-slope hermetic roof, among other elements (Fig. 18).

5.2.2 Weakening of localisation path

The variation in the degree of localisation among cluster B-1, cluster B-2, and cluster B-3 likewise reflects differing localisation strategies adopted in response to the design imperatives of distinct historical periods in modern China. Over time, these divergent strategic choices have given rise to a discernible path in which the localisation of foreign public buildings within cluster B, representing a classical-localised process, gradually weakened from the 1840s to the 1930s.

In the 1840s, due to a variety of practical and institutional constraints, many foreigners who arrived in Shanghai did not immediately construct Western-style residences. Instead, they resided in existing Chinese-style dwellings and adapted to the local built environment. This early period of cross-cultural encounter was marked not only by a passive acceptance of traditional Chinese cultural influences but also by a gradual process of cultural identification and selective acceptance. Notably, Westerners, particularly those engaged in missionary activities, began to adopt architectural forms that were already familiar to the local populace, which facilitated communication and minimised resistance. Religious buildings, in particular, often incorporated traditional Chinese architectural features (Zheng, 2020), which formed a deliberate and distinctive expression of diversified architectural localisation during this historical period.

The distinguished British historian Toynbee (1987) observed that, in the face of encounters with foreign or heterogeneous cultures, a process of cultural compromise, which he termed the “principle of moderation”, often served as a transitional mechanism that resolved conflict and fostered adaptation. In the case of the mid-19th-century Jiangnan region, the initial collision between Western and Chinese cultural paradigms did not result in a wholesale rejection of local architectural practices. Instead, under the influence of the principle of moderation, the design of early foreign buildings in Shanghai continued to rely heavily on indigenous construction techniques and stylistic conventions.

As a result, many early architectural examples exhibited clear features of localisation. The tendency of such diversified architectural localisation was most prominently observed in the improved vernacular buildings categorised within cluster B-3, such as the Yangjiaqiao Catholic Church in Suzhou, built in 1892, and the Xinlan Bookstore in Wenzhou, built in 1910. These buildings displayed localised characteristics in multiple formal dimensions, including material structure, facade modelling, and decorative patterns. For example, the continued use of traditional brick (stone) wood hybrid structure, familiar roof lines, and indigenous decorative patterns exemplified a conscious negotiation between foreign and native forms, which offered compelling evidence of early architectural hybridity shaped by cultural pragmatism and historical context. These influences continue to shape architectural approaches in the region today.

However, around the beginning of the 20th century, various unequal treatments in modern China’s politics, commerce and trade influenced the tendency towards westernisation in the ideology of modern Chinese society, and the application of localised design showed a tendency to weaken. The ideological bias of worshipping the West and devaluing China had a biased impact on the application of the local design approach and the continuation of the local design thinking of the modern Jiangnan foreign architecture. ‘Zheng Jiguo Zhuyi’ (anti-patriotism) surged before the 1920s. Qian (1914) described that anti-patriotism at that time viewed European and American countries as the best, from their rules and regulations to their clothes, ambitions, and hobbies. Everything Chinese was tainted with a sense of disdain and described as unreliable. Therefore, classical architectural styles and design elements associated with the leisure classes in Europe and the United States were extensively adopted in China as progressive and anti-traditional symbols of modern capital. This facilitated the widespread dissemination of Western classical architectural prototypes during the 1920s, subsequently influencing the facade expressions of foreign public buildings in Jiangnan in modern times, which can be seen, for instance, in the Southeast Asian colonial architecture-style facade of the Wuxi County Library and the Eurasian architecture-style facade of Hangzhou Sicheng Hall in cluster B-2.

Even so, as the construction technology of foreign public buildings during this period had not yet fully diverged from the traditional architectural practices of Jiangnan, their material structures and decorative patterns still retained many local characteristics. According to research, the earliest use of a fully reinforced concrete frame structure in Jiangnan occurred with the construction of the Shanghai Telephone Company in 1908. This technology was subsequently adopted more broadly in high-rise buildings, which were particularly those in the Art Deco style, in cities such as Shanghai, Tianjin, and Wuhan from 1913 through the 1930s (Li, 2004). Nevertheless, due to constraints related to materials, technology and funding, reinforced concrete frame structures were not widely implemented in foreign buildings in small and medium-sized cities. Consequently, structures such as the Wuxi County Library (a Southeast Asian colonial-style building) built in 1911, and the Hangzhou Sicheng Hall (a Eurasian-style building) constructed in 1927, continued to employ traditional Chinese brick (or stone) and timber hybrid structures as their primary framework. The decorative detailing of these buildings also largely adhered to traditional Chinese decorative pattern themes.

It is worth noting that the emergence of cluster B-1 indicates that, after 1930, a distinctive building type developed―one that diverged significantly from the prevailing trend of localisation at the time. This architectural form exhibited a relatively weak degree of localisation and represented a transitional phase in the evolution from Western classical architectural prototypes to modern architecture. Despite a growing emphasis on local characteristics after 1930, many architects, particularly foreign ones, continued to associate Western architectural forms with notions of progress, authority, and modernity. For instance, the Suzhou University Gym (Romanesque architecture) in cluster B-1, completed in 1937 with funding from the Southern Methodist Church, retained continuous Romanesque arches on its facade. It only reflected traditional Chinese architectural characteristics in the contour of its roof.

The diminishing adoption of localised architectural forms may not have been purely an aesthetic decision; rather, it likely had ideological underpinnings. It may have served as a manifestation of a Western-dominated narrative of modernity (Kryvoruchko, 2019). In the eyes of these architects, the continued application of the Romanesque architecture style extended their religious values and cultural mission. It also symbolised their “modernising mission” through the use of Western classical architectural language in a foreign context, influenced by Western modern architectural thought. Such ideology is particularly reflected in the public buildings of Suzhou University, as the Suzhou University Gym, which combines education and religion.

However, such classically rooted forms gradually lost their appeal with the rise of Western modernist thought. Moreover, the global economic exhaustion brought about by the Second World War (1939—1945) further shifted architectural trends (Moser, 2015). Therefore, after the 1940s, the architectural language of modern public buildings in the Jiangnan region largely returned to Vernacular revival architecture expressions and Western modernist simplicity. An example is the former site of the Former National Resources Commission of the Nationalist Government in Nanjing (Vernacular revival architecture) in cluster B-1, constructed in 1947, which reflected a revivalist local style. From this point forward, architectural expression tended to favour practicality, simplicity, and one-dimensional localised forms.

6 Conclusion

This study examined how Western architectural stylistic prototypes were localised in Jiangnan’s public buildings during the modern period of China (1840—1949). Using a multidimensional clustering method, it presented an information technology approach to apply quantitative analysis to architectural history, trying to form a new non-architectural perspective on how classical and modern Western styles were adapted within a Chinese context.

Through data coding and decoding, two distinct paths for the localisation of the formal characteristics of foreign public buildings in modern Jiangnan emerged: (1) cluster A represented the ‘modern-localised’ type. Over time, the localisation of architectural formal characteristics intensified, evolving from single-aspect design elements such as decorative patterns to a multi-aspect approach encompassing both facade modelling and decorative patterns. This trend reflected a progressive deepening in the degree of localisation. (2) cluster B represented the ‘classical-localised’ type. In contrast, the localisation of architectural features in this group diminished over time. Initially characterised by a comprehensive design involving material structures, facade modelling, and decorative patterns, it gradually regressed to focus solely on facade modelling. This trend reflected an overall decline in the degree of localisation.

In light of the historical context surrounding the construction of foreign buildings in modern China, the two evolutionary paths of localisation in the formal characteristics of public buildings in the Jiangnan region revealed that localised design expression was not static, but rather a process of continuous adaptation in response to the evolving needs of the time. Among these, the differences in localised design also reflected architects’ in-depth thinking about the relationship between locality and foreignness in architectural design before and after modern China, which provided important inspiration for contemporary architects to address the relationship between globalisation and regionalisation in regional architectural design. In the 1950s, Sicheng Liang put forward the Theory of Translatability of Architecture, believing that by replacing the compositional elements of a Western-style building with corresponding Chinese elements, the building could be transformed into a Chinese style (Lai, 2009), emphasising the role of bridging the localised architectural language with the internationalised architectural context―a process similar to the changes in the localised characteristics of stylistic prototypes in Jiangnan’s foreign public buildings in recent times. In the future, regional architectural stylistic design can further focus on the dynamic characteristics of architectural localisation and pay attention to the trend of applying localised characteristics in multiple design dimensions such as plan modelling, facade modelling, material structure and decorative patterns, so that they are naturally transformed and adapted to different eras and cultural contexts, thereby constructing a Chinese architectural stylistic design discourse system with regional characteristics and significance for the times.

Although multidimensional data clustering provided a possible and systematic quantitative method for analysing the localisation of architectural stylistic prototypes, several limitations remained. The findings should be regarded as preliminary, outlining only the general paths of stylistic prototype localisation. This research therefore represented an early stage in the quantitative study of architectural styling transformation in the Jiangnan region. Future studies could expand the dataset and employ broader empirical validation to refine and substantiate the observed trends in the evolution of modern foreign-influenced architecture in Jiangnan, thereby enhancing the rigour and comprehensiveness of the conclusions.

References

[1] Bodart, A.M., 2003. Eclectic Architecture. University of Cincinnati, Cincinnati.

[2] Chen, X., 2017. Neoteric architecture style influenced by industrial and commercial culture―survey based on Suzhou. In: Wuxi and Changzhou. Jiangnan University, Wuxi.

[3] Cheng, K., Neisch, P., 2023. A new perspective on eclectic attributes in architecture: taking eclectic architecture in Beijing and Hong Kong as an example. J. Asian Architect. Build Eng. 22 (3), 1126—1145.

[4] Corbett, G.G., 2012. Features. Cambridge University Press.

[5] Cui, K., 2007. 2007-10. Following in the footsteps of Mr Liang Sicheng to learn on the path of architectural localisation―speech at the 2007 academic annual meeting of the architectural Society of China. In: Proceedings of the 2007 Annual Academic Conference of the Architectural Society of China, Xi’an.

[6] Dai, M., 2018. The Study on the Modern Early Modernist Style Architecture in Jiangnan Region. Jiangnan University, Wuxi.

[7] Dreyer, C., 2010. Architectural Codes from a Semiotic Perspective. Code: Between Operation and Narration, pp. 55—73.

[8] Dursun, P., 2012. Dialogue on space: spacial codes and language of space. A|Z ITU J. Fac. Archit. 9 (1), 104—119.

[9] Edwards, A.T., 2013. Architectural Style. Read Books Ltd. Redditch.

[10] Eisenman, P., 2004. Digital scrambler: from index to codex. Perspecta 35, 40—53.

[11] El-Mahdy, D., 2025. From theory to practice: truchet tile as a computational visual coding approach for facade visualization. Int. J. Architect. Comput. 23 (2), 464—480.

[12] Gleiniger, A., Vrachliotis, G., 2012. Code: Between Operation and Narration. Walter de Gruyter, Boston.

[13] Guo, W., Luo, J., 2015. Nantong Modern ‘East Meets West’ Architecture. Southeast University Press, Nanjing.

[14] Gura, J., 2015. The Guide to Period Styles for Interiors: from the 17th Century to the Present. Bloomsbury Publishing USA, New York.

[15] He, B., 2023. Research on Clustering Algorithm of Multidimensional Spatial Data Complexity Features. Harbin University of Science and Technology, Harbin.

[16] He, J., 2017. Principles of Urban and Rural Planning. China Agricultural University Press, Beijing.

[17] Hu, Z., Liu, P., 2020. Memory and Expression of Homeland―Theory and Practice of Traditional Settlement Landscape Mapping. The Commercial Press, Beijing.

[18] Kadir, T.A.Q.R.A., Kassim, P.S.J., Majid, N.H.A., Latip, N.A., 2019. Themes of classicality in the Malay architectural form: principles from an aristocratic realm. Melayu: Jurnal Antarabangsa Dunia Melayu 12 (1), 27—59.

[19] Kryvoruchko, Y., 2019. Innovativeness of the Tradition of Religious Architecture. IOP Conference Series: Materials Science and Engineering, Bristol, England.

[20] Lai, D., 2009. Architectural Translations and the Notion of Architectural Translatability in Modern China the Architect, vol. 1, pp. 22—30.

[21] Lai, l., Tong, Y., 2017. Building Mechanics and Structures. Beijing Institute of Technology Press, Beijing.

[22] Legendre, P., Legendre, L., 2012. Numerical Ecology. Elsevier. Radarweg.

[23] Li, H., 2004. Zhongguo Jian Zhu Xian Dai Zhuan Xing. Southeast University, Nanjing.

[24] Li, Z., Zhang, H., Sun, Y., Geng, Z., 2015. Anomaly Detection Algorithms in Non-numerical Data. Jiangxi People’s Publishing House. Nanchang.

[25] Liu, J., 2006. Chongan Mingsheng Shihua. Shandong Pictorial Publishing House, Jinan.

[26] Luke, H., Wu, C., 1936. ‘Women De Zhuzhang (Our Proposition)’. the Chinese Architect, vol. 26, pp. 55—56.

[27] Ma, L., Zhang, T., Zhou, D., 2023. From division of labor, professional collaboration to multi-work collaboration: Research on the operation mode of parallel buildings in the context of digital technology. New Arch. 6, 106—111.

[28] Miles, M., 2013. Metabolism: a Japanese modernism. Cult. Polit. 9 (1), 70—85.

[29] Moser, J.E., 2015. Global Great Depression and the Coming of World War II. Routledge, London.

[30] Pan, G., 2009. A History of Chinese Architecture. China Architecture & Building Press, Beijing.

[31] Qi, K., 2000. Architecture, space and Morphology―on the architecture morphology. J. Southeast Univ. (Nat. Sci. Ed.) 30 (1), 1—9.

[32] Qian, Z., 1914. Zheng Jiguo Zhuyi (Anti-patriotism). Orient. Advert. (4), 1—4.

[33] Qu, W., Liu, T., Zhang, L., Wang, H., 2011. The Design and Analysis of Algorithms. Tsinghua University Press, Beijing.

[34] Rowe, P.G., Kuan, S., 2002. Architectural Encounters with Essence and Form in Modern China. Mit Press, Cambridge.

[35] Shalunts, G., Haxhimusa, Y., Sablatnig, R., 2011. Architectural style classification of building facade windows. International Symposium on Visual Computing. Las Vegas.

[36] Tan, P., Steinbach, M., Kumar, V., 2006. Introduction to Data Mining. Posts & Telecom Press, Beijing.

[37] Tatarkiewicz, W., 1980. A History of Six Ideas: an Essay in Aesthetics. Springer, Netherlands. Heidelberg.

[38] Toynbee, A.J., 1987. A Study of History: Volume I: Abridgement of Volumes I-VI, vol. 1. Oxford Paperbacks, New York.

[39] Turchin, P., 2018. Historical Dynamics: Why States Rise and Fall. Princeton University Press, Princeton.

[40] Wei, Z., Guo, Q., Lu, C., Yang, L., Ying, T., 2021. Shape similarity measurement based on DNA alignment for buildings with multiple orthogonal features. Acta Geod. Cartogr. Sinica 50 (12), 1683—1693.

[41] Xia, B., Li, X., Shi, H., Chen, S., Chen, J., 2020. Style classification and prediction of residential buildings based on machine learning. J. Asian Architect. Build Eng. 19 (6), 714—730.

[42] Xia, Z., 1999. Ci Hai. Shanghai Lexicographical Publishing House, Shanghai.

[43] Yue, J., Hu, G., 2024. A study of English rococo style under the influence of “Chinese Style”. Chin. Trad. Cult. 12 (1), 67.

[44] Zhang, G., 2006. Map of China’s Architectural Culture. China Environmental Science Press, Beijing.

[45] Zheng, D., 2004. On the heated controversyhe departmental culture and we sternization in the 30’s journal of social science of. Hunan Norm. Univ. 33 (3), 84—90.

[46] Zheng, S., 2020. The Evolution of Shanghai Architecture in Modern Times. Tongji University Press, Shanghai.

Rights & permissions

2095-2635/2025 The Authors. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd.