1. Introduction
Rapid urbanization in recent decades has led to a significant increase in urban populations, generating substantial demand for ecological, recreational, and shopping experiences [
1,
2]. Places in cities that provide ecological, recreational, and shopping functions are usually well developed [
3], either dense and vibrant with commercial and public facilities, attracting a large number of human activities [
4,
5], or offering comfortable ecological spaces that allow for access to nature, promoting residents’ physical health [
6] and mental health, and improving their sense of wellbeing [
7]. These places often possess the unique character and charm of the city and are highly representative of it [
8,
9]. We can call these Representative places (RPs).
Representative places (RPs) is not a concept widely recognized by the academic community. Kevin Lynch lists five features for identifying cities in his work
Urban Imagery, including nodes, landmarks, edges, paths, and zones [
10]. The city can be represented and viewed through popular locations within these elements [
11,
12]. Therefore, Representative places (RPs) can be surface-like spaces such as commercial districts and tourist areas, linear spaces such as characteristic commercial streets and riverfront scenic zones, and point-like locations such as landmarks and urban nodes. However, this functional richness, as well as spatial diversity, makes RPs a collection of multiple research objects, posing a challenge to RP research. Nevertheless, since RPs play an important role in enhancing the local cultural perceptions of tourists and improving the life experiences of citizens [
13], it is crucial to focus on RPs.
Identifying RPs and defining their spatial boundaries has become one of the primary challenges in the study of RPs, as urban research, urban management, environmental restoration, planning, and design all require a clear understanding of an object’s extent. Existing place identification studies have utilized appropriate data sources and effective tools to perform identification from various perspectives. To enrich the semantics of places, researchers have used point-of-interest and social media data to identify places and their functions and types. This is usually achieved by analyzing density or quantity and is synchronized with the identification of the places’ scope [
14,
15]. Social media platforms such as Twitter, Weibo, and Flickr provide many explicit geographic coordinate points that are generated by the public and are semantically rich [
16,
17]. However, these coordinate points often suffer from location bias, as users usually prefer famous attractions or popular places for check-ins, even if they are not actually there [
18]. To obtain more accurate check-in locations, some studies have employed data-assisted methods. For instance, to account for location bias, J. Huang et al. [
19] concatenated geotagged text and image data from Twitter and Instagram with survey data. POI data are the second most commonly used data source. Because it carries rich functional semantics, researchers use semantic information to identify the functional space of places. For example, Pan et al. [
20] used POI data to identify the neighborhood functional areas of Macau Peninsula, Yang et al. [
21] delineated the functional areas of the city by stacking buildings with POI data, and K. Liu et al. [
14] identified the cognitive boundaries of transportation nodes using POI data. It is evident that most of these studies apply data to large-scale analyses of faceted urban functional areas, with less focus on the point and line spaces of small-scale places. In terms of spatial boundary identification methods, polygonal algorithms [
22] and kernel density estimations [
23] are commonly used to identify the ranges. Hobel et al. [
24] employ natural language processing and machine learning techniques for identification, while K. Liu et al. [
14] utilize DBSCAN clustering and Chi-shape algorithms, among other techniques, to generate cognitive boundaries for metro stations. Although these studies make useful attempts to address boundary ambiguities, they do not effectively address perceptually based boundary recognition. Furthermore, even when boundary ranges are extracted, issues such as lack of accuracy and irregular shapes persist.
In summary, to accurately identify places and delineate the scope of place perception, it is essential to select data that accurately reflects human perception of places and to find innovative ways to associate people with places. Since RPs are multifactorial collections, there have been no direct studies on RPs; researchers have predominantly focused on environmental spaces. Numerous studies have examined human cognition and perception of the environment through the interaction of the sensory organs with the environment, encompassing five aspects: hearing, touch, smell, taste, and vision. Auditory research focuses on the urban acoustic environment, particularly on how urban sounds impact human psychology and emotion [
25,
26,
27]. Research in haptics (thermal comfort) focuses on skin contact with thermal environments, particularly on how these environments affect human health risks [
28,
29]. Additionally, more studies are focusing on the visually impaired, examining their tactile perception of crosswalks [
30,
31]. The term “smellscape” is frequently used to describe how the sense of smell, spanning across spaces, significantly impacts human behavior, attitudes, and health [
32]. However, it has rarely been addressed or measured in planning practice and theory so far. Recent efforts have focused on understanding and regulating environmental odors, particularly on detecting, controlling, and eliminating maladaptive odors [
33,
34]. Taste research in urban environments is also relatively scarce at present. Existing studies primarily focus on the types of food offered by restaurants and ordered by users, examining the distribution of tastes in these environments [
35,
36] and the taste behaviors of the population [
37]. Vision is the primary focus of built environment studies, which are mainly conducted through visual images, such as streetscape pictures, green landscapes [
38,
39], land use [
40], and transportation spaces [
41]. In particular, the development of machine learning and deep learning technologies has enabled large-scale, high-resolution, and multi-dimensional perception studies of built environments [
42,
43]. In summary, the research on built environments based on human perception is very large and fruitful. The relevant results are mostly from the visual perception level, and the research on hearing, touch, smell, taste, and so on is relatively small, which is mainly due to the difficulty of obtaining data sources. This is mainly due to the difficulty of obtaining data sources. So far, the acquisition of large-scale and high-granularity data sources has not been well solved. Existing visual perception research methods are relatively mature. Street view images are a location-based quality data source that provides a data base for place-based visual perception research. Deep learning-based human perception of street view images greatly facilitates visual perception quantitative research. Although visual perception research is more fruitful, less attention has been paid to comparative studies of specific places in cities, and there is also a lack of systematic visual perception evaluation of these places.
To address these issues, this study proposes a novel approach to identify and evaluate RPs (RPs) in cities from a human perception perspective. Our approach integrates multi-source data, including POIs, street view images, and social media, using techniques such as deep learning and text semantic analysis. Initially, popular check-in locations were extracted from microblog check-in data. Based on the matching pattern between these popular locations and POI names, a cohesion method that combines spatial clustering and polygon generation techniques was proposed to identify RPs and to extract the perceptual regions (PRRPs) of RPs. Subsequently, using the six dimensions of human perception as assessment indicators, the human perception characteristics of PRRPs were examined in terms of different types, levels, and preferences of various groups of people, and the perceptual differences between RPs and non-RPs were evaluated. We will address two key scientific questions: (1) How can RPs (RPs) and their Perceptual Ranges (PRRPs) be identified in terms of human perception? (2) How do human perceptions at RPs vary across types, levels, and population preferences? By answering these questions, this study aims to deepen our overall understanding of Representative places in the city, provide valuable insights into placemaking, management, and sustainability in the city, and provide guidelines for urban planners and designers in place planning and design.
5. Discussion and Conclusions
5.1. The RPs and Their Perceptual Regions
This study introduces an innovative approach to identify RPs and their perceived areas within the city through POIs combined with social media data. Our results identify 192 RPs in Nanjing. Of these, 89 are Parks, 71 are Streets, and 32 are Squares. Park-type RPs were the most common. On the one hand, Nanjing is surrounded by mountains and water, and the abundant natural resources within its borders reflect the characteristics of the city of Nanjing. On the other hand, it aligns well with the development intentions of the Nanjing Municipal Government. Nanjing proposes that by 2035, the coverage rate of green areas and public spaces larger than 400 square meters within a five-minute walking distance in built-up areas will reach more than 95%. With the significant increase in urban parks and green spaces, human perceptions of space tend to favor the Park type.
We categorized the RPs into three levels: 11 level 1 places, 27 level 2 places, and 120 level 3 places. The first level consists of highly recognized places that serve as windows to the city of Nanjing, but all of these places are located in the main city of Jiangnan. Additionally, the number of secondary and tertiary places in Jiangbei is far less than that in Jiangnan Main City. The Nanjing Territorial Spatial Master Plan (2021–2035) (Draft) proposes enhancing the functions and quality of the new main city in Jiangbei. Therefore, in the future, Nanjing needs to select qualified places for cultivation among the secondary and tertiary places in the main city of Jiangbei. This will help improve the functions of the new main city of Jiangbei by adding one or two primary places and several secondary places.
We identified four categories of places: visitors like and local residents dislike, local residents like and visitors dislike, both local residents and visitors dislike, and both local residents and visitors like. The number of places in each category was 11, 12, 13, and 22, respectively. In addition to the findings from the conclusions, we found that places popular with both visitors and local residents usually have a larger range of place perceptions. It is logical that the more popular an RP is, the more likely it is that the various facilities around it will be named after the RP, making it more memorable and recognizable. These findings can provide insights for subsequent RP remodeling and planning.
5.2. Human Perception of the RPs
RPs can be the city’s transport hubs, business districts, or well-known attractions, and providing high-quality RPs is crucial to enhancing the city’s image. We used six dimensions of human perception as assessment metrics for evaluating RPs. Our study reveals spatial differences in the perception of RPs. Overall, each perception indicator in the RP perception area shows better performance than in the non-RP perception area, indicating positive spatial effects within the RP perception area. This finding fully reflects the inherent representativeness of these places. For RPs with a small percentage of hot and cold spot grids, spatial optimization can be performed in conjunction with their area and identification. Previous studies have discussed the relevance of the six dimensions to streetscape elements. Therefore, the human perception score can be improved by adding more relevant streetscape elements.
5.3. Potential Applications and Research Limitations
RPs have the city’s unique personality and charm and best represent its character. Understanding the spatial distribution and human perception of RPs is essential to enhance visitors’ cultural perceptions and citizens’ life experiences. Firstly, RPs are identified by the popularity of visits, and the scope of the place is determined by people’s spatial perception, which truly reflects their willingness. Armed with this knowledge, urban planners and designers will be able to optimize or plan for different types and levels of places that meet the preferences of visitors and local residents, thus forming a comprehensive, functional, and hierarchical network of functions. Second, the human-perceived evaluation of RPs encompasses the human evaluation of the environment and facilities. Through these evaluations, it is possible to grasp people’s overall impression of the city at the macro level; at the meso level, it is possible to formulate guidelines for remodeling places of different types, grades, and preferences of different groups of people; and at the micro level, it is possible to put forward detailed remodeling suggestions for specific places. Compared to traditional small-scale analyses, the use of large amounts of social media and SVI data can cover the entire city and quickly complete the identification and assessment of RPs. This street-image-based approach offers advantages in terms of efficiency, cost-effectiveness, and time saving compared to a field survey of the same scale. However, it is important to acknowledge the limitations of this study. Firstly, social media data are limited to microblogging data, which is skewed and may exclude some population segments, resulting in biased place identification results [
57]. For example, some older adults are also important populations for RP use, but they generally do not publish Weibo, so targeted supplementation of missing population data is necessary. Potential measures for this are increased questionnaires for missing groups and statistics on useful information about missing people in conjunction with attraction traffic data. Secondly, the static nature of the Baidu street view platform’s images hinders the acquisition of historical images, thus ignoring the impact of seasonal changes on human perception. In vegetation-rich places, there may be significant differences in landscape characteristics between winter and summer [
58]. Therefore, methods such as crowdsourcing can be used to supplement missing and anachronistic streetscapes based on the time of the study. Thirdly, due to the limitation of the training dataset, we only used six dimensions of human perception as evaluation factors. Methods such as crowdsourcing can be used to add more targeted perceptual evaluation factors. Fourthly, the training dataset contains many European and North American cities, but not Nanjing, which could lead to potential perceptual biases due to changes in the urban environment. This can be accomplished by censoring out data with very different architectural styles and adding data for the target cities. Fifth, ResNet50 is not state-of-the-art, although it is a more effective neural network model. ResNet50 was chosen because of the low training data. Deeper, more advanced models require more data, a limitation that can be improved with subsequent increases in training data.
Due to space constraints, this paper does not discuss the influencing factors that form the spatial pattern of RPs. An in-depth discussion of the influencing factors can reveal Nanjing’s historical development and current economic pattern, thus providing more practical solutions for future urban development. In addition, based on the findings of the current RPs identification and assessment, a more complex analysis of the interactions between residents and tourists in the urban space can be carried out by combining the survey data or interviews, and the related findings can further support urban management, built environment renovation, as well as sustainable urban development. The above limitations can be gradually resolved in further studies in the future.