A bibliometric review of perceptions of the built environment supported by multi-source data

ABSTRACT

Information and digital technologies have revolutionized research methodologies for measuring people’s perceptions of built environments. Vast amounts of urban data can now be collected, providing more accurate and intuitive presentation and analysis tools. This enhances the research on the built environment from a human-centered perspective. In this paper, with the help of the CiteSpace knowledge map analysis tool, manual deep intervention was combined with clustering and stratified sampling of literature information to identify hot topics and research trends. The results demonstrate that there are two new perspectives. The first is the up-bottom urban/group perspective, specifically a holistic study of group perceptions by large-scale data, represented by big data, with technologies such as artificial intelligence. The second is the bottom-up humanistic/individual perspective, which accurately examines individual perceptions through physiological data based on ergonomics techniques, psychological, and other small-scale data. Furthermore, these two perspectives are interconnected, indicating a cross-trend in both directions. By bridging the gap between subjective perception and the real-world environment, integrating multi-source data offers a more precise and efficient foundation for thoroughly evaluating the built environment, which would help us better understand the built environment and satisfy the users’ demands in the research and practice of urban design.

Graphical abstract

KEYWORDS:

• Literature review
• built environment
• perceptions
• integration of multi-source data
• CiteSpace

1. Introduction

The built environment is a multidisciplinary research area that includes architecture, planning, geography, and transportation. It is a complex system composed of elements related to land use, transportation systems, and urban design (Frank and Engelke 2001; Handy et al. 2002; Saelens and Handy 2008). “How do people perceive and use urban space, and to what extent and in what ways does the objective environment influence human perception and behavior?” is one of the central themes of the discipline of architecture. Environmental perception is the psychological environment generated by people after receiving and processing information in the physical space environment, and it can guide behaviors (Gold 1980).

Research on perceptions of the built environment as a reflection of functional-oriented modernist urban planning appeared very early and has developed from classical studies to 5D Theory. Pioneers such as Jacobs and Lefebvre began to focus on human-scale space and its social and economic effects in the 1960s (Jacobs 1961; Lefebvre 1962), and scholars such as Jan Gehl, William Whyte, and Kevin Lynch constantly summarize how to improve the vitality of urban space (Gehl 1971; Lynch 1984; Whyte 1980). Then, the measurement of built environment perception is mainly based on the classical 5D Theory of “density, diversity, design, destination accessibility, distance to transition” (Cervero and Kockelman 1997; Cervero et al. 2009), and a relatively wide range of studies have been carried out.

While the classical studies are still instructive for current urban research, there is a lack of in-depth study supported by quantitative data and evaluation framework. Moreover, some scholars have discovered that people’s perceptions of the built environment may also affect their physical and mental health (Roe 2008; Xu et al. 2019). This 5D Theory cannot cover these factors, making human perception evaluation challenging to consider. Perception measures rely on data, and due to the complexity of the built environment itself, related research and theoretical explorations have been characterized by separate approaches. Compared to objective environmental measures of physical space, the innovative explorations of perception measures in existing studies need further exploration (Clarke et al. 2008; Cunningham-Myrie et al. 2015; Michael et al. 2006).

New technologies such as artificial intelligence (AI), virtual reality (VR), augmented reality (AR), mixed reality (MR), and wearable physiological sensors bring new opportunities for perception measures of the built environment. On this basis, with the help of Building Information Modeling (BIM), Geographic Information System (GIS), and other visualization platforms, the concept of Digital Twin (DT) in the Architecture, Engineering, Construction, and Operation (AECO) sector has become a reality (Piras, Agostinelli, and Muzi 2024). From this, the methods to measure perceptions of the built environment have been continuously developed and innovated. The basic idea of the current research is how to measure people’s perceptions of their surroundings more comprehensively, accurately, and conveniently. This makes it easier to explore the interrelationship between people’s perception and the environment supported by various perceptual technologies and approaches. In this background, topics on new research highlights, research hotspots, and research trends are worth further discussion. In the background mentioned above, the number of publications has grown rapidly and received wide attention, but the development of the overall research still needs to be sorted out. Therefore, this paper is organized around three questions: “What stages of development has the research field in perceptions of the built environment gone through?” “What types of perception data and methods of perception measures are available?” “What research hotspots have arisen, and what are the research trends?” The research objectives are derived from the answers to the above questions. Firstly, the overview of the research field is organized to identify the main academic subjects involved and to understand the intrinsic implications of the study further. Secondly, by summarizing the data types and measure methods, methodological and technical references are provided for evaluating urban spatial qualities from a human-centered perspective. Thirdly, through the exploration of research hotspots and trends, the theory of refined urban design and renewal is enriched.

2. Materials and methods

2.1. Data sources

Applying bibliometric analysis to the perceptions of built environment measurement study, this paper takes the following steps:

1. Selection of the literature database. This study searches and downloads the Web of Science Core Collection (WoS) data to ensure that the data are comprehensive, accurate, and powerfully explanatory. WoS is an authoritative and scientific literature database widely used globally. It contains most of the prestigious and influential journals in the field of the built environment, with sufficient and updated data. CiteSpace is a tool developed by Prof. Chaomei Chen in 2003 based on Java language for drawing knowledge maps, which is still continuously being updated and improved. Through in-depth data analysis of co-cited literature and references, CiteSpace presents related fields’ knowledge structure and development trends in colorful maps, realizing multivariate, time-sharing, and dynamic visual analysis of a research field (Chen 2006). Meanwhile, it has good compatibility with the WoS database，which is selected as the primary data source in this study. “Built Environment” and “Humanistic Perception” are the two themes that focus on.

2. Determination of the search keywords and filtering scope. Measuring built environment perception involves the concepts of built environment and human perception. The former means the environment of urban built-up areas (such as public spaces, streets, squares, etc.), and the latter means the interaction between humans and the built environment, emphasizing the subjective experience of the urban spatial elements formed through processing the human cognitive system.

This paper selects “Topic” as the search type to retrieve title, abstract, author, and keywords from the WoS core set. Based on the balance of phrases, the TS= (“built environment” OR “urban space* “OR “urban form*” OR “urban morphology* “OR “urban street*” OR square* OR “outdoor environment*” OR “public space*“OR “living environment*”) AND (audit* OR assess* OR evaluate*) AND (space perception* OR human perception*), To cover all relevant studies as much as possible, there is no restriction on the literature type and time range in searching. Considering the evidence-based approach, the ergonomics experiment equipment and method, the design orientation of urban renewal in the new data and new technology background, the additional TS is = (“augmented reality*” OR EEG* OR eye-tracking* OR “eye movement*” OR electroencephalogram* OR “virtual reality*” OR “machine learning*” OR “physical sensor*” OR GSR* OR “galvanic skin response*”)AND(“urban design” OR “urban renewal”).

Revising and replacing the irrelevant research keywords and TS, the search results were repeatedly screened manually, and 780 literature records were obtained. Then, the result is imported into CiteSpace, and the number of literature records was reduced to 588 after excluding proceedings papers and reviews. Finally, these records were selected as the data source for this study. The date of data search and download was 18 July 2023.

2.2. Research methods

This paper made an overall study of the perception measures of the built environment through knowledge maps and qualitative analysis. Typical methods of literature review include qualitative reviews and systematic reviews with meta-analyses. The former covers extensive issues but is more subjective, while the latter relies on quantitative statistics and strict selection criteria. With data and information visualization development, CiteSpace has become one of the most popular and representative tools for knowledge graphs. This study combined these two methods to make a comprehensive literature analysis. We drew a scientific knowledge graph with the help of CiteSpace, an information visualization and analysis software. Then, we read the critical literature according to qualitative review and finally visualized the development process.

In the Citespace V6.2.R3 software, we mainly select Keywords, Categories, References, etc., and use the algorithm of pathfinding and network slicing to conduct multivariate clustering analysis on the research field of human-centered perceptions of the built environment. Next, we use the knowledge graph, visual analysis, and statistical tables to collect data such as the number of publications, subject categories, countries, and keywords. The results were sorted and cited for evidence analysis. Meanwhile, the core literature is based on the traditional qualitative review to clarify the research development tracks, the change of thinking mode, and follow-up research hotspots (Figure 1).

Figure 1. Outline of perceptions of the built environment (PBE) research design.

3. Results

3.1. Results by data collection

3.1.1. Quantity of publications

The change in research publication over time can reflect the situation of the study in a certain period, which is an important indicator to evaluate the development of this research field. Figure 2 illustrates the information on the date and the number of all publications related to “perceptions of the built environment”. There are three stages of research development.

Figure 2. Annual distribution of research literature on PBE,2002–2023.

Stage 1 - Exploration (2002–2006): The three research themes of “built environment”, “humanism”, and “perceptions” received independent attention from scholars of different subjects in the last century. However, few studies combined these three research themes to influence human cognition before 2006. The initial stage of this research field is set from 2002 to 2006.

Stage 2 - Growth (2007–2016): The data change in this period shows a relatively large fluctuation but a rise through the overall trend. There are increases of 71% from 2008 to 2009 and 27% from 2014 to 2015, respectively, which indicates that new researchers began to contribute to this field. However, there is little systematic understanding of this research area during this stage because of the significant fluctuation.

Stage 3 - Development (2017-present): The average annual growth rate of published documents from 2018 to 2021 is 53%, with a small number of fluctuations but an overall increasing state. After 2019, the number of published papers every year is more than 50, and it rise to 128 in 2022. The topic of “perceptions of the built environment” has gradually gained attention from global scholars, and there is a tendency to become an academic hotspot.

3.1.2. Distribution of regional cooperation

Analysis of the literature published by different countries can reflect their diverse contributions and influence in this field (Chen et al. 2012). The top 10 countries in terms of the proportion of the number of published literature are China (22.1%), the United States (21.6%), the United Kingdom (9.5%), Germany (7.7%), Canada (6.1%), Australia (5.3%), Japan (4.9%), Italy (4.6%), the Republic of Korea (4.6%) and Brazil (4.3%) (Table 1). CiteSpace was used for co-occurrence analysis of regional literature in this research field. The node type was “country”, the time slice was one year, and the visual analysis was based on the number of published papers. The outer ring of the node circle is purple, indicating that the node has a high centrality, which reflects the academic cooperation and influence in this region. Figure 3 shows the countries where research in this field has a high impact worldwide, such as the United States, the United Kingdom, Sweden, and France. The highest centrality of the publication is in the United Kingdom, which shows scholars are actively participating in academic cooperation and exchange with others in other nations.

Figure 3. Knowledge map of regional cooperation distribution in PBE, 2002–2023.

Table 1. List of contributing countries/regions and number of records contributed in PBE, 2002–2023.

No.	Freq	Centrality	Year	Country/Region
1	130	0.04	2003	China
2	127	0.14	2003	USA
3	56	0.33	2002	England
4	45	0.08	2002	Germany
5	36	0.06	2004	Canda
6	31	0.09	2007	Australia
7	29	0.02	2007	Japan
8	27	0.07	2004	Italy
9	27	0.01	2007	South Korea
10	25	0.03	2002	Brazil

For the emergence of research from different countries or regions, the UK (2002–2008) and Canada (2004–2011) can be the predecessor. Studies in both countries lasted for over seven years. The image intensity value can indicate that research in this field has been hot for a long time in both countries and shows a high amount of publications (Figure 4). The emergence intensity in a country/region, characterized by the emergence length and strength, can reflect the geographical distribution of major studies in the relevant field over a certain period (Chen and Min 2019; Kexin et al. 2023). From the perspective of emergence intensity, the research on the “perception measures of the built environment” presents a longer time and higher value of emergence in the northern hemisphere countries. We can find that the research results in the Northern Hemisphere countries are more than those in the Southern, and they receive more attention from the global academic community.

Figure 4. Top 10 countries/regions with the strongest citation bursts in PBE, 2002–2023.

3.1.3. Structure of subject knowledge

The research has involved 242 subject categories internationally. By summarizing the information on multidisciplinary cooperation and analyzing the knowledge structure, we can find the hot subject change and interdisciplinary collaboration in the past 20 years. Table 2 illustrates the top 3 subjects based on the publication volume: Neurosciences (84), Environmental Sciences (51), and Environmental Studies (48). In addition, some disciplines with high centrality (≥0.2), such as Ergonomics (0.58), Environmental Sciences (0.33), Public Environmental& Occupational Health (0.37), Radiology, Nuclear Medicine and Medical Imaging (0.2), Geography, Physical (0.24), Computer Science, Artificial Intelligence (0.33), reflect their strong connection with other disciplines in the research process of built environment perception measurement, which serve as multidisciplinary linkages to promote interaction between disciplines.

Table 2. List of the most influential subject categories in PBE, 2002–2023.

No.	Freq	Centrality	Year	WoS Categories
1	84	0.19	2002	Neurosciences
2	51	0.33	2012	Environmental Sciences
3	48	0.17	2004	Environmental Studies
4	43	0.31	2003	Engineering, Electrical & Electronic
5	40	0.11	2007	Construction &Building Technology
6	32	0.04	2011	Engineering, Civil
7	30	0.03	2017	Computer Science, Information Systems
8	29	0	2009	Green & Sustainable Science & Technology
9	29	0.13	2004	Urban Studies
10	29	0.37	2003	Public Environmental & Occupational Health

The literature co-occurrence map (Figure 5) of the research fields generated by CiteSpace can visualize the relationship between the major subjects. This study consists of three main directions: First is centered on Neurosciences, including Construction & Building Technology, Green & Sustainable Science & Technology, Engineering, Clinical Neurology, and Electrical & Electronic Clinical Neurology, etc.; Second is based on Ergonomics, which covers Behavioral Sciences, Ophthalmology, Radiology, Nuclear Medicine & Medical Imaging, and others. Third is Environmental Studies, which links to various subjects such as Urban Studies, Computer Science, Information Systems, and Public Environmental & Occupational Health. The co-occurrence map demonstrates a strong correlation and close combination between the research directions of Neurosciences and Ergonomics (Figure 5), and it maintains active academic progress in their related research fields (Bower, Hill, and Enticott 2023; Brielmann et al. 2022; Latini et al. 2023; Y. Zhang et al. 2022). The interdisciplinary research related to Environmental Studies deserves further attention.

Figure 5. Co-occurrence map of subjects categories in PBE, 2002–2023.

According to the emergence analysis results of CiteSpace (Figure 6), Neuroscience was a prominent subject in early research (2002) on “perception measures of the built environment”. It maintained hot academic discussions in the following 10 years, which indicated that Neurosciences was once a hot subject in this field, showing a relatively high emergence value overall. Then, Radiology, Nuclear Medicine, Clinical Neurology, and Ophthalmology emerged, promoting research fever in this field in these subjects lasting for at least five years. With the research improving over the years, this field has accumulated research results in more disciplines.

Figure 6. Top 10 subject categories with the strongest citation bursts in PBE, 2002–2023.

3.2. Research trends

If two or more literature become references in one or more papers later, they will build a co-citation relationship (Chen et al. 2012). Co-citation is a highly recognized scientific phenomenon. Based on this principle, Citespace incorporates individual citation literature into a collection of citation literature databases and sorts out the co-citation network, generating co-citation clusters, which form the knowledge base in this field (Small 1973). With the help of CiteSpace’s literature co-citation analysis function and cluster analysis function, key literature in this field can be easily identified to understand the research trend better. On this basis, the literature co-citation network and traditional qualitative review can also provide further analysis for this research through the overall lens and subdivision topics, which results in a comprehensive understanding of the human-oriented perception measurement field of the built environment.

3.2.1. Outline of research

The study uses a one-year time slice (January 2002 to June 2023) and keeps the top 25 high-frequency cited literature records in each time slice following the CiteSpace Guide. Figure 7 illustrates the field’s co-citation analysis map of literature, including 681 nodes with 1338 links. To further understand the outline of the research field, we selected the top 20 high-frequency cited articles for qualitative analysis, which provided this study with a more in-depth view to help learn about the whole condition (the core topics discussed) of this field.

Figure 7. Knowledge map of literature co-citation analysis in PBE, 2002–2023.

According to these 20 articles, the current research focuses on large-scale measurements supported by new urban data (Street View Images represented) and new technology (Machine Learning represented). Street view images integrating Convolutional Neural Networks (CNN) image recognition technology have gradually become a mainstream research method. It works as a “medium” to combine traditional design theories and investigative methods with other technologies, such as computer technology and biotechnology. Rondel et al. (2011) have proved that streetscape evaluation is coherent with subjective investigation. At present, pertinent research focusing on infrastructure assessment and management has been used in the study of built environment in majority of cities worldwide such as green space distribution and quality assessment (X. Li, Yang Cai, and Ratti 2018; X. Li et al. 2015; Ye et al. 2019), spatial order and community health (Bader et al. 2015; F. Liu and Kang 2018; Qiu et al. 2021; R. Wang et al. 2019; Zhou et al. 2019), visual perception assessment (Ma et al. 2021; Yao et al. 2019; F. Zhang et al. 2018), dynamic assessment of visual quality (Tang and Long 2019), and audio-visual spatiotemporal perception mapping (Sanchez et al. 2017; Verma, Jana, and Ramamritham 2020). In this context, Scholars have proposed several related ideas, including walkability (Nagata et al. 2020; Zhao et al. 2017), green visual index (X. Li et al. 2015; Lu 2019; Ye et al. 2019), openness (X. Li, Carlo, and Ian 2017; X. Li, Yang Cai, and Ratti 2018), etc.

In the MIT Media Lab project “Place Pulse”, the crowdsourcing method is used to enable visitors to evaluate the quality differences between two street view images through an online website (http://pulse.media.mit.edu), which helps to achieve a large-scale, fine-grained evaluation of urban streetscapes. In Place Pulse 2.0, the indicator of quality extends to safety, vitality, wealth, and scenic beauty (Chester et al. 2015; X. Li et al. 2015; Naik et al. 2014; The PLOS ONE Staff 2015; Qiu et al. 2021; F. Zhang et al. 2018).

3.2.2. Subfields of research

Generally, the literature co-cited frequently relates to the research’s concept, theory, or methodology. Clustering algorithm analysis in CiteSpace enables these documents to build a literature collection based on the co-citation intensity, which is the clustering, then the co-citation network composed of multiple clusters further reflects the growing trend and cutting-edge area of a subject or knowledge field during a specific period (Chen 2006; Chen, Ibekwe‐SanJuan, and Hou 2010). Latent Semantic Indexing(LSI), Log-Likelihood Rate (LLR), and Mutual Information (MI) are three different algorithms provided by CiteSpace for extracting clustering labels. The LLR and MI algorithms highlight the research characteristics, while the LLR algorithm is more practical and less repetitive (Kexin et al. 2023). Two indicators, the modularity Q and mean silhouette, are often used to assess the reasonableness of clustering. When modularity Q is greater than 0.3, the clustering structure is proved to be significant; when the mean silhouette is greater than 0.5, the clustering is reasonable, and when it is greater than 0.7, it is convincing (Chen et al. 2012).

According to the process suggested in the CiteSpace Guide, we use the LLR algorithm to get three sub-clusters in automatic clustering from the cluster analysis. The modularity Q is 0.9472, and the mean silhouette is 0.9668. The Knowledge Graph(Figure 8) thus can reveal a remarkable cluster structure and reasonable clustering mode (Chen, Ibekwe‐SanJuan, and Hou 2010). Next, the quantitative analysis of the co-citation of literature can further identify three subdivided topic clusters (#0 objective measure, #1 reducing stress, and #2 public open space). The new topic clusters were summarized and reorganized after in-depth reading and qualitative analysis of each cluster’s top 10 high-frequency cited articles.

Figure 8. Cluster map of literature co-citation in PBE, 2002–2023.

Overall, the research mainly focuses on street-oriented urban public space (#0 objective measure, L. Wang et al. 2022; Wu et al. 2021). Around 2015, with the emergence of big data such as street view images and point of interest (POI), studies with both large-scale and fine-grained characteristics were realized, and the data put more emphasis on multi-source (Chester et al. 2015; X. Li et al. 2015; The PLOS ONE Staff 2015). Apart from traditional methods such as on-site investigation and questionnaire interviews, the three-dimensional spatial analysis method that combines street view image data and image recognition technology has further supported the research and realized off-site and automated group perception measurement in the urban spatial scale (#0 objective measure, refer to Chapter 3.2.1). From 2015 to 2020, physiological research topics related to landscape environments, such as stress recovery, attention recovery, and emotional perception, have become research hotspots (#1 relieving stress). The research methods focus on synthesizing subjective psychological feelings and objective physiological manifestations. Moreover, technology or tools such as virtual reality and physiological sensors have enabled quantitative analysis of the built environment to present a new development direction through a human-centered perspective.

Under the lockdown policy caused by COVID-19 in recent years, new modes of working, living, and socializing have changed people’s behavior patterns and needs. Considering the importance of daily physical and mental health, people have raised higher demands on the living space and natural environment (Y. Liu et al. 2023). New technologies represented by EEG and eye movement have realized the objective and accurate acquisition of the real-time physiological feelings of the population., The collected source data effectively supports the research on improving the quality of the built environment, which contributes to alleviating physiological and psychological problems caused by the spatial environment for urban residents (Bower, Hill, and Enticott 2023; Brielmann et al. 2022; Chamilothori et al. 2022; Eloy et al. 2023; Guo et al. 2022; Latini et al. 2023; C. Li, Ge, and Wang 2022; Liao et al. 2022; Y. Zhang et al. 2022).

3.2.3. Research hotspots

Keywords are the core ideas that highly summarize the main content, academic concepts, research objectives, and methods of the paper, and high-frequency keywords can reflect the hot research areas. Using CiteSpace to generate a knowledge map of keyword co-occurrence, we can find that the keywords with the highest frequency of occurrence are perception, virtual reality, machine learning, design, and intermediary centrality are all very high, which to a certain extent can reflect the hot areas in the research (Figure 09).

Figure 9. Knowledge map of keyword co-occurrence in PBE, 2002–2023.

To further clarify the distribution of hot topics of built environment perception and the co-occurrence relationship of research topics. Following the process suggested in the CiteSpace guidelines, we obtained a map consisting of 588 nodes and 1546 links with 10 meaningful keyword clustering modules in a time slice of one year (January 2002-June 2023). The mean silhouette of the cluster is 0.8616, and the clustering modularity Q is 0.7201.This results in the acquisition of 10 significant keyword clustering modules, including deep learning (#0), EEG (#1), heart rate variability (#2), artificial neural network (#3), eye movements (#4), and pain perception (#5), depressive symptoms (#6), virtual reality (#7), auditory oddball (#8), outdoor thermal comfort (#9) (Figure 10). After further summarization, it can be found that the 10 clusters can still form 2 key research directions: (1) “Group perceptions supported by large-scale data and new technology at city scale” (#0, #3) and (2) “Individual perceptions supported by small-scale data and new technology at humanistic scale” (#1, #2, #4, #5, #6, #7, #8, #9).

Figure 10. Cluster map of keyword co-occurrence in PBE, 2002–2023.

Research hotspots are often represented by CiteSpace’s burst feature (Nayak et al. 2018). In Figure 11, “cortex,” “attention,” and “eye movements” are the most prominent keywords, with an average of six years of popularity. Since the 2010s, research methods have gradually broadened, and new technologies and methods, such as support vector machines and independent component analysis, which emerged in 2015, have received continuous attention. Additionally, they focus more on the impact of human behavior, such as user acceptance in 2017. In 2021, the terms “view”, “scale”, and “preference” mainly relate to the quantitative study and problem diagnosis for a particular aspect or several elements of the built environment, based on relevant data and technical methodologies, along with optimization plans. Until now, the research has remained hot. Meanwhile, with the improvement of computer arithmetic power and the continuous development of digital technology, how to combine with artificial intelligence technology to presume that the scientific and refinement of the human-centered built environment will be the focus of research in the next few years (Brielmann et al. 2022; Eloy et al. 2023; Latini et al. 2023; Liao et al. 2022; Miao, Wang, and Li 2021; Tang and Long 2019; Ye et al. 2019; F. Zhang et al. 2018).

Figure 11. Top 20 keywords with the strongest citation bursts in PBE, 2002–2023.

4. Discussion and conclusion

4.1. Discussion

4.1.1. Types of multi-source data

The development of information technology has provided new data, methods, and technologies for built environment research. Urban research has changed from traditional mechanical thinking to more open and forward-looking data thinking. Multi-scale urban data also provide a more comprehensive and systematic research idea for perceptions of the built environment (Table 3).

Table 3. Types and applications of multi-source data.

Data scale	Data type	Representative study	Application Scenario	Disadvantages
Large-scale	Street View images, such as Google, Tencent, Baidu, and others	Li et al. (2015); Zhang et al.(2018); Ye et al.(2019); Tang and Long (2019)	Image labeling and automatic classification are used with crowdsourcing and computer vision techniques to identify street scene aspects and quantify the spatiotemporal change analysis of street scenes.	Involving personal privacy and public information security risks, some data has limitations.
	Service facility data, such as Point of Interest (POI), public transportation data	Junfeng et al.(2018); Pang and Zhang (2015); Quercia et al.(2014); De et al.(2016); Miao et al. (2021)	The functional diversity and prevailing function of the urban environment can be examined.
	Social media data, such as Twitter, Facebook, Sina Weibo data		Activity behavior, place preference, and regional recognition are examined by combining semantic word frequency analysis and other methods with the user’s public space activities or text data, including geographical coordinates.
	Mobile location data, such as mobile signaling, GPS, WIFI probes, location-based service (LBS)		Urban vitality and diversity correlation analysis, activity density, large-scale and large-sample behavior pattern, demand identification, and simulation.
Small-scale	Subjective questionnaire and interview data	Lynch (1984); Gehl (1971)	Small-scale examination of traditional subjective perception	Influenced subjectively by the “perceptual threshold”
	Physiological data, such as eye tracking, EEG, skin electrocardiogram	Ulrich et al. (1991); Birenboim et al.(2019); Bakir-Demir et al.(2021)	Real-time, objective, and accurate recording of people’s emotional and physiological changes in the urban environment	Differences in algorithms and no unified standards

4.1.1.1. Large-scale data for urban/group perceptions

Large-scale data reflect the spatial and temporal characteristics of group perceptions, providing the possibility of rapidly measuring refined spatial patterns and capturing spatial perceptions of residents on a large scale. There are now two primary categories : (1) Based on mobile location data (GPS, WiFi probes, location-based services, LBS, etc.), social network data (Twitter, Sina Weibo, etc.), and visual representation of human behavior, behavior, activity regulations, etc., has good real-time and extensive (Jiao et al. 2017; Pang and Zhang 2015; Quercia, Schifanella, and Maria Aiello 2014); (2) Data that has been refined using information from the Street View images, service facility data like public comments, mobile signaling (De Nadai et al. 2016), Points of Interest (POI), and public transportation data, etc. These data can be quantitatively analyzed by combining them with Geographic Information System (GIS) to meet the dual requirements of large-scale and high-precision perceptual measurements (Miao, Wang, and Li 2021). Compared to traditional data, large-scale data has disadvantages such as bias, sample error, and personal privacy disclosure.

4.1.1.2. Small-scale data for humanistic/individual perceptions

Individual-scale perceptual data accurately identify personal perceptions and are obtained in two main ways: subjective perceptual and physiological index measures. The Subjective Perception Scale is the most widely used emotion measurement tool. Others include the Disgust Sensitivity Scale (DSS), the Multidimensional State Boredom Scale (MSBS), the State-Anxiety Inventory (SAI), the Positive and Negative Affect Scale (PANAS), the Profile of Mood States (POMS), and the Perceived Restorative Scale (PRS). Individual subjective tendencies, weather, time of day, and other factors easily influence the analytical results of subjective perception scales. However, they are still the widely applied method in built environment perception research.

An essential addition to objective data is physiological data, which can better identify the impact of the environment on people’s feelings, even those who are not directly aware of the environmental feelings (Bakir-Demir, Kazak Berument, and Akkaya 2021; Birenboim et al. 2019). Ulrich et al. (1991) conducted the first empirical examination of the systematic use of physiological markers to quantify spatial perception.

4.1.2. Classification of research methods

Typical measures of the built environment perception are broadly categorized into three types: group perception supported by big data and artificial intelligence technology, individual on-site perception in real-world environments, and individual virtual perception in laboratory environments (Table 4).

Table 4. Classification of research methods.

Scale	Type	Measure method	Representative study	Advantages and disadvantages	Application Scenario
Urban/Group （up-bottom）	Group perceptions supported by big data and artificial intelligence technology	Image segmentation technology based on different algorithms, such as Convolutional Neural Networks (CNN),support vector machines, random forest, etc.	Zhang et al.(2018); Rossetti et al.(2019); Verma et al.(2019); Li et al.(2018); Ye et al. (2019); Tang and Long (2019); Li et al.(2015); Lu (2019)	With the help of machine learning and other technologies to achieve efficient processing of large-scale data, it is difficult to reproduce all the information of the urban environment as a whole and data refinement is limited.	Comparative analysis of several cities and urban environment research at the municipal level
Humanistic/ Individual （bottom-up）	Individual on-site perceptions in real-world environments	Subjective evaluation scale, semi-structured interview, behavioral observation,physiological wearable sensors (such as eye movement, EEG, skin electrocution), etc.	Ewing and Handy (2009); Lynch (1984); Bakir-Demir et al.(2021); Gehl (1971); Mehta (2014); Zamanifard et al.(2019); Lucas and Romice (2010); Patrik and Stigsdotter (2010)	Perception data from real-world environments are reliable, but the test scope is small and the workload is large.	The analysis of specific urban spaces, such as streets, squares, etc.
	Individual virtual perceptions in laboratory environments	Virtual reality technology, augmented reality (AR), subjective perceptual measure, physiological wearable sensors, such as eye movement, EEG, skin electricity, etc.	Mauss and Robinson (2009); Karakas and Yildiz (2020); Brielmann et al.(2022); Jeon and In Jo (2020); Sun et al.(2019); Zhang et al.(2021); Schlickman (2019); Pia (2019); Liu et al.(2020); Ren and Kang (2015); Latini et al.(2023)	The test is simple and the variables are controllable, but the experimental scenes are difficult to reproduce all the information of the real-world environmens.	Specific studies of the effects of single or multiple variables on perception in a laboratory simulation environment

4.1.2.1. Group perception supported by big data and artificial intelligence technology

Large-scale group perceptions are quantitatively characterized, taking advantage of the recent global emerging new data environments combined with intelligent recognition, image segmentation, and other machine learning datasets created by perception crowdsourcing. The intervention of image recognition modeling can significantly improve the efficiency and accuracy of cities in terms of management and prediction, and the research mainly focuses on urban landscape analysis, problem detection, and urban assessment (F. Zhang et al. 2018; Rossetti et al. 2019; Verma, Jana, and Ramamritham 2019).

4.1.2.2. Individual on-site perception in real-world environments

The process of measuring on-site perception involves the researcher choosing real-world environments, setting up participants for wandering experiences, and finishing up the collection of physiological signals and on-site multisensory perceptions (Ewing and Handy 2009; Lynch 1984; Bakir-Demir, Kazak Berument, and Akkaya 2021; Gehl 1971; Mehta 2014; Zamanifard et al. 2019; Lucas and Ombretta 2010; Grahn and Stigsdotter 2010). The main advantage of on-site perception measures is the authenticity and comprehensiveness of perception, while the disadvantage is that it is more challenging to control additional environmental variables strictly, and there are limitations in sample size, bias, and representativeness.

4.1.2.3. Individual virtual perception in laboratory environments

In order to collect physiological signal data from subjects, determine the correlation between one or more variables and the target emotion, and effectively eliminate uncontrollable factors, virtual perceptions involve re-presenting and regulating real scenes in the laboratory with the aid of virtual reality (VR) and other technologies (Mauss and Robinson 2009; Karakas and Yildiz 2020; Brielmann et al. 2022; Jeon and Jo 2020; Sun et al. 2019). Contrary to the on-site perception approach, the applicability of the experimental results to rich and complex urban spaces leaves much to be desired. In the meantime, most presentations of virtual perception – like those by Zhang et al. (2021), Schlickman (2019), and Fricker (2019) – are visual. Recent landscape evaluation studies have highlighted the value of multi-sensory experiences, citing the connection between soundscape and the natural environment (Y. Liu, Hu, and Zhao 2020; Ren and Kang 2015) and the workplace (Latini et al. 2023).

5. Conclusion

With the help of CiteSpace Knowledge Map, co-citation analysis, and keyword clustering, the overall 20 high-frequency cited articles and the top 10 articles in each subfield were qualitatively analyzed, and the hotspot researches were sorted out. However, due to the limitation of human and material resources, only a single database of WoS was used as the source of primary data, which is a deficiency of this study.

In the currently rising trend of human-centered planning and the rapid development of information technology, perceptions of the built environment have gradually become more important. Multi-modal data such as sensor data, social media data, mobile device data, street view image data, etc., are gradually becoming new data to supplement or replace traditional data. At the same time, the help of intelligent algorithms, artificial intelligence, virtual reality, wearable physiological sensors, and other digital technologies to analyze and understand urban space has become an irreversible trend nowadays. By summarizing the literature on perceptions of the built environment, the main subjects, including Neurosciences, Environmental Sciences, and Environmental Studies, show a multidisciplinary cross-trend.The developmental stages of the study were sorted out by identifying three main periods: exploration (2002–2006), development (2007–2016), and growth (2017 to the present), with a brief summary of each stage. Meanwhile, the data types and classifications, the methods of measurement and their application scenarios are summarized in tables (Tables 3 and 4), which support the construction of a systematic research methodology for perception measures of the built environment from a human-centered perspective.

There are two main perspectives: firstly, “up-bottom” group perception measures supported by large-scale data and artificial intelligence technologies, and secondly, “bottom-up” individual perception measures supported by small-scale data, such as psychological and physiological data, etc. At the same time, we should be wary of the blind worship of methods and data, and avoid overemphasizing quantitative research to the detriment of perceiving and observing environmental characteristics and practical problems.The three main methods for measures of the built environment perception are group perception supported by big data and artificial intelligence technology, individual on-site perception in real-world environments, and individual virtual perception in laboratory environments. Humanistic/individual perception data is small-scale, and its combination with large-scale data can support group perception assessment and drive research on the mechanisms influencing behavior and perception of the built environments.

In future research, the integration of multi-scale data will become one of the hotspots, focusing on the effects between the objective characteristics and subjective perception to answer the fundamental question of “to what extent and in what way does the physical and spatial environment affect people’s feelings and behaviors” more accurately. It also opens up the possibility of reforming the research paradigm to support the “human-environment interaction”. In engineering projects, new data and technologies can be integrated into the urban design process, and the results of perception measures can be used to form a quantitative design basis and evaluation criteria. We can grasp people’s perceptions of the built environment in a comprehensive and integrated way from the “up-bottom” group and the “bottom-up” individual perspectives and further promote the conscientization and refinement of urban design and regeneration practices.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The work was supported by the National Natural Science Foundation of China [52078113].

Notes on contributors

Tingjin Wu

Tingjin Wu, a Ph.D candidate in Department of Landscape Architecture, School of Architecture, Southeast University, China. His research interests include evaluation of urban spatial perception and campus landscape.

Jinxiu Wu

Jinxiu Wu, professor in School of Architecture of Southeast University, China, mainly engaged in the research of green building design and renovation, as well as integrated design of architecture and landscape.

Yanxiang Yang

Yanxiang Yang, is a Ph.D candidate in Landscape Architecture at Southeast University, China. His research interests include low-carbon campus, urban data analysis and digital design.

Wei Dong

Wei Dong, is a Ph.D candidate in Landscape Architecture at Southeast University, China.