Associations Between Environmental Factors and Perceived Density of Residents in High-Density Residential Built Environment in Mountainous Cities—A Case Study of Chongqing Central Urban Area, China

Abstract

In high-density built environments, perceived density (PD)—shaped by physical, socio-cultural, and perceptual factors—often induces sensations of crowding, stress, and spatial oppression. Although green spaces are recognised for their stress-reducing effects, the influence of built-environment characteristics on public sentiment under stringent mobility restrictions remains inadequately explored. This study takes Chongqing, a representative mountainous metropolis in China, as a case to examine how natural and built environmental elements modulate emotional valence across varying PD levels. Using housing data (n = 4865) and geotagged Weibo posts (n = 120,319) collected during the 2022 lockdown, we constructed a PD-sensitive sentiment dictionary and applied Python’s Jieba package and natural language processing (NLP) techniques to analyse emotional scores related to PD. Spatial and bivariate autocorrelation analyses revealed clustered patterns of sentiment distribution and their association with physical density. Using entropy weighting, building density and floor area ratio were integrated to classify residential built environments (RBEs) into five tiers based on natural breaks. Key factors influencing positive sentiment across PD groups were identified through Pearson correlation heatmaps and OLS regression. Three main findings emerged: (1) Although higher-PD areas yielded a greater volume of positive sentiment expressions, they exhibited lower diversity and intensity compared to low-PD areas, suggesting inferior emotional quality; (2) Environmental and socio-cultural factors showed limited effects on sentiment in low-PD areas, whereas medium- and high-PD areas benefited from a significantly enhanced cumulative effect through the integration of socio-cultural amenities and transportation facilities—however, this positive correlation reversed at the highest level (RBE 5); (3) The model explained 20.3% of the variance in positive sentiment, with spatial autocorrelation effectively controlled. These findings offer nuanced insights into the nonlinear mechanisms linking urban form and emotional well-being in high-density mountainous settings, providing theoretical and practical guidance for emotion-sensitive urban planning.

1. Introduction

Chongqing, a representative high-density mountainous metropolis in China, faced significant challenges in controlling the spread of infectious diseases within its central urban areas during the 2022 outbreak. According to a report from the Chongqing Channel of People’s Daily Online (http://www.people.com.cn/, (accessed on 28 August 2025)), over 60% of confirmed cases in the core metropolitan region resided within the same building, community, or neighbourhood [1]. This clustering highlights the heightened risk of transmission in densely populated built environments, where residents spend prolonged periods in high-intensity enclosed spaces—a condition that may exacerbate physical and psychological stress.

Although urban green spaces, public activity areas, and essential service facilities such as markets, convenience stores, and pharmacies have been shown to support emotional regulation and alleviate mental distress [2,3,4], community interactions during such periods also posed a dual risk: facilitating both social support and potential epidemic spread. This tension underscores the complex trade-off between management efficacy and mental well-being in high-density urban settings.

Against this backdrop, social media platforms such as Weibo have emerged as vital channels for information sharing, emotional expression, and collective support [5,6]. User-generated content offers valuable insights into public sentiment, revealing nuanced perceptions and needs during crises.

1.1. Research Advances in Perceptual Density

1.1.1. Definition of Perceived Density

The qualitative enhancement of the urban built environment is a pivotal concern in modern urban development in China. The announcement of the Healthy China 2030 Plan [7] asks the government to urgently enhance public health by optimising urban planning, advancing infrastructure development, and elevating the quality of public services. The proliferation of high-density urban residential buildings (RBs) engenders significant urban health challenges in downtown areas [8]. Nevertheless, urban density includes both physical density and PD; most current scholars concentrate on the physical density of urban areas, and PD is neglected. Research has revealed that the human impression of space frequently holds greater significance than objective spatial indicators. According to Amos & Rapoport [9], PD is defined as an individual’s subjective assessment of the population density, available space, and spatial arrangement within a specific location. Meanwhile, individuals evaluate spatial density in relation to social standards and usually attribute the importance of PD, especially in high-density environments, to its crucial role in crowding. ‘Crowding’ is typically defined as a negative subjective assessment of excessive apparent density, coupled with psychological stress [10].

1.1.2. PD Measurement

Currently, research on PD exhibits a pattern of numerous approaches coexisting, each offering distinct insights but revealing inherent limitations. The laboratory research conducted in environmental psychology employs psychophysical measuring tools, such as the Likert scale and semantic differential approach, to assess perceived crowding through a controlled image-based visual perception study. This method can effectively isolate the perception mechanism; however, its artificial environment often fails to replicate the intricate density experience influenced by various sensory cues, including auditory, olfactory, and thermal sensations, in a genuine urban context. To compensate for this deficiency, the field survey in urban design incorporates technologies such as wearable sensors (e.g., eye movement tracking) for on-site measurement. While ecological validity has been enhanced, it encounters logistical obstacles in large-scale data collecting and frequently overlooks the brain underpinnings of density perception. Advancements in computational modelling technology, particularly methods utilising CNN street view analysis and LBS mobile data analysis, present novel opportunities for extensive PD research; however, these data-driven approaches frequently overlook the evaluation process and essential cognitive socio-cultural factors, such as the crowding tolerance threshold in cultural mediation. This methodological gap has generated a comprehensive framework that amalgamates perceptual immediacy and cognitive evaluation, highlighting that PD arises not solely from sensory input but also through multi-tiered interpretations influenced by personal and collective experiences. These studies mainly collect data on site [11], Nevertheless, field studies are labour-intensive and costly and pose significant challenges for data collection at large scales.

The intersection of disciplines reveals the inherent trade-offs of research methodology, underscoring the significance of a mixed-methodologies approach. Cognitive neuroscience can precisely identify neural mechanisms but lacks ecological context; landscape ecology measures pro-natural effects but fails to fully account for sociocultural influences; urban anthropology benefits from ethnographic insights but struggles to translate ritualised spatial usage into a predictive model. The text-based analysis method offers a novel approach for elucidating the socio-cultural formation of density cognition. Through the analysis of geo-tagged social media data using NLP technologies, researchers may ascertain the correlation between emotional load descriptors and particular metropolitan attributes. In addition, Hofstede’s cultural dimension theory suggests that collectivist societies are more likely to see density as social vitality instead of personal discomfort [12]. This amalgamation of computational and qualitative methodologies not only encapsulates the immediate physiological reaction elicited by density but also elucidates the cultural importance of influencing enduring views.

Affective methodologies utilising social media data to analyse public discourse and integrating environmental psychology with computational and cultural frameworks facilitate the identification of critical areas for service enhancement and bolster enduring public health surveillance strategies while effectively merging benefits and mitigating temporal and spatial constraints [13]. In parallel, regression models—ranging from Ordinary Least Squares (OLS) and spatial syntax to machine learning, deep learning, and random forest algorithms—have become mainstream tools for analysing the influence of diverse environmental and social factors on public perceptions of stress [14,15].

The prevalent use of mobile devices and social media to document and disseminate mood states presents an opportunity to collect extensive data for monitoring public health trends and assessing perceptions of the built environment in densely populated urban areas [16,17,18,19,20].

1.2. Environmental Factors in High-Density Residential Built Environments of Mountainous Cities

Natural and built environment factors: The intricate terrain and geomorphology of blue-green zones in mountainous cities profoundly influenced their initial urban design and cultivated a rich cultural history. Urbanisation has imposed land limitations, while the growth of high-rise structures has transformed the city skyline and encroached upon ecologically sensitive regions. This has resulted in diminished happiness and well-being among people, especially in waterfront areas [21]. Changes in height difference, the clarity of the field of vision, and the edge effect of the waterfront have a big impact on how open and oppressive a room feels. For instance, doing things by the sea can help relieve mental stress and make you feel less anxious and depressed [22]. The density of traffic nodes and the shape of the road network in the built environment are closely tied to how many people are concentrated in one place and how fast they move, which influences how easily residents can get to their homes and how quickly they can travel. Parks and natural attractions, as significant open spaces, mitigate the psychological stress associated with high-density developed settings by offering visual greenery and recreational options [23]. Many studies have shown that going into green areas to do things and making parks easier to get to can boost citizens’ physical and mental health [24,25].

Physical density factors: Physical density in residential surroundings profoundly influences people’s perceptions of density. Physical density encompasses geographical metrics such as floor area ratio (FAR), building density (BD), and height, in addition to social density, which includes population density and difficult-to-quantify qualitative factors. Woodbridge et al. [26] discovered that increasing residential density adversely affects residents’ psychological well-being, whereas Van Dyck et al. [27] established a negative correlation between high density and neighbourhood satisfaction. Additional research indicates that these impacts may exhibit nonlinear characteristics [28].

Socio-cultural factors: Socio-cultural elements, such as environmental identification, cultural norms, and the accessibility of social services, significantly influence outcomes. In high-density mountainous urban regions, the concentration of intensive spaces amplifies residents’ perceived need. Improving the safety of walkways and the accessibility of recreational facilities in green spaces and pocket parks might enhance resident happiness [29].

1.3. Associations Between Sentiments of PD and Environmental Factors in RBEs

As a significant psychological measure of felt stress, sentiment monitoring has become more important post-pandemic. Traditional methods collect physiological data from surveys to assess public environmental views. The amount and breadth of precise data gathering are limited, and the temporal perspective is modest, making innovative discoveries difficult. With the rise of digital communication platforms, large-scale sentiment analysis using NLP on social media data is useful in urban planning and public health research [30].

By analyzing user-generated content—such as comments and posts—researchers can uncover spatial and temporal patterns in collective sentiment, offering insights into public responses to environmental conditions and urban interventions. Despite its growing utility, several methodological challenges remain. One common issue is the misalignment between the geographic metadata of social media posts (e.g., from platforms like Weibo) and the actual locations described within the text, leading to spatial inaccuracies. Furthermore, sentiment classification models may fail to associate sentiments with specific environmental features or planning elements relevant to the study context. To address these challenges, keyword extraction using word segmentation systems can help identify text directly related to the research target, thereby narrowing the semantic scope and enhancing spatial specificity [31]. Advances in deep learning models have substantially improved the precision of sentiment classification. In contexts with limited annotated data, manual validation remains a viable approach to increase model reliability and ensure contextual accuracy. Additionally, statistical methods provide a structured foundation for preliminary analysis and large-scale data interpretation. Spatial grid-based clustering techniques further support the integration of environmental attributes, enabling categorical aggregation and distributional analysis [32]. These methods serve as the groundwork for more complex spatial modelling and inform evidence-based urban policy and management decisions. Understanding the determinants of PD across various high-density urban settings is vital for improving residents’ well-being and optimising urban environments. Although the Chongqing Land and Space Master Plan (2021–2035) outlines broad development guidelines for high-density areas, it lacks detailed strategies for the spatial allocation of service facilities. Therefore, it is essential to classify Chongqing’s RBEs and conduct PD analysis across multiple spatial scales to inform more targeted and responsive urban interventions.

2. Materials

2.1. Case Study Area

Located in southwest China, Chongqing (28°10′ N−32°13′ N, 105°17′ E−110°11′ E) is in the eastern Sichuan Basin and covers 4779 square kilometres in a folded parallel ridge-valley. The parallel ridge-valley, which consists of mountains with a continuous orientation and parallel boundaries, spans between 10 and 30 km in width. The alternately parallel mountains and valleys, steep mountain terrain, and elevations under 1000 m make it a distinctive mountainous landform [33] (Figure 1). Despite its proximity to the Yangtze and Jialing River basins, the city has few inland water bodies. Water features are few in high-density locations, while waterfront spaces are abundant in low-density areas. Natural limits have created a high-density urban development pattern that affects daily life. Chongqing City’s urban form is multi-centre and cluster-based due to nature and other causes. Each area can constitute a comparatively independent place with relatively complete regional amenities and production and living activities in this organic urban design. Thus, examining the inherent link between environmental elements in residential daily activity areas and inhabitants’ PD is crucial.

Chongqing’s urban area was mostly in Yuzhong District before the current city was built. The 2023 population of Chongqing was 31.9143 million, ranking first among China’s four direct-controlled municipalities [34]. As population grows and socioeconomic improvement continues, over 90% of Yuzhong District, the primary urban region, has a FAR of 5.0 or above [35]. However, many slowly developing suburbs on the fringes of the main urban region require policy improvements adapted to varied development demands. Regional studies like the one in the central urban area of Chongqing, a high-density city with worldwide natural environmental characteristics, are crucial.

2.2. Assessment Framework

Given the intricacy and multifaceted nature of urban living, shown by Chongqing during the pandemic, high-density cities require a more sophisticated hierarchical framework to analyse the correlation between perceived density and environmental circumstances. We also hypothesise that the impact of emotional factors on perceived density may vary considerably among residential contexts with differing levels of high density. The framework’s detailed hierarchical structure aids researchers and planners in comprehending the contextual relationship between perceived density and emotional well-being, differentiating environmental factors across various density gradients, and facilitating targeted urban policies that reconcile public crisis and control with mental health in high-density settings (Figure 2).

2.3. Data Preprocessing

Weibo text, housing data were scraped via Python 3.9. Service facilities POI, water and road data, precipitation, humidity, temperature and 30 m resolution Digital Elevation Model (DEM) data were downloaded from official website. Data description is shown in

Table 1. Description of the data sources in this study.

Types	Description	Data Sources	Content
Weibo text	SINA Weibo	https://weibo.com/ (accessed on 10 June 2024)	Coordinates in the WGS-84 coordinate system, review, ID
Housing data	Fang Holdings Limited	https://cq.esf.fang.com/https://weibo.com/ (accessed on on 11 March 2024)	Coordinates in the WGS-84 coordinate system, FAR, building type, number of households, residential type, floor area
Service facilities POI	Baidu Map	https://map.baidu.com/https://weibo.com/ (accessed on on 21 January 2024)	4 categories including 12 elements
Water and road data	Open Street Map	https://www.openstreetmap.ie/https://weibo.com/ (accessed on 25 January 2024)	Area features, line features
Climate Data	Chinese Academy of Sciences Resources and Environment Cloud Platform	https://www.resdc.cn/https://weibo.com/ (accessed on 25 August 2025)	Coordinates in the WGS-84 coordinate system, precipitation, humidity, temperature
30 m Digital high-range model (DEM)	Geospatial Data Cloud	http://www.gscloud.cn/https://weibo.com/ (accessed on 12 January 2024)	Line features

. All data were projected to the GCS WGS 1984 coordinate system to ensure consistency of the basemap.

2.4. Data Cleaning

After obtaining the Weibo data in CUA from 1 January 2022 to 31 December 2022, the Weibo data needed to be cleaned. First, I deleted the data in the Weibo dataset that contained only blank text, including images and videos. Second, HTML tags and URLs have been stripped, and emojis and special characters have been standardised.

A large amount of tourist data would affect the accuracy of sentimental analysis in residential areas. We refer to the research of Qu and Girardin et al. [36,37] and use the feature engineering and classification algorithm in machine learning to identify tourists and residents. The process includes: (1) Model training. In the Chongqing super topic, all comments in 2022 from 100 IDs, including 50 residents and 50 tourists, were manually selected, and the text data was randomly allocated according to a 20% training set and an 80% test set. (2) Operation: Take N in “N consecutive days” as a variable and try different values (such as 15 days, 20 days, 30 days, 40 days, 50 days, and 60 days). The features include ‘Total number of posts’, ‘Number of active days’, ‘Maximum number of consecutive active days’, ‘Active time range (days)’, ‘Average number of posts per day’, ‘Posting frequency’, ‘Number of recent active days’, ‘Standard deviation of posting time’, and ‘Consecutive active days exceeding N’, among which the active time range (days) was the most significant (

Table A1. Distribution of feature importance.

Feature	Significance
Active time days	0.23123
Standard deviation of posting times	0.20661
Posting frequency	0.16354
Active days	0.13408
Total posts	0.12820
Days of recent activity	0.07953
Maximum consecutive active days	0.05191
Average daily posts	0.00491

).

Table A2. Distinctive characteristics yield differing sensitivity training outcomes.

Strategy	Accuracy	Precision	Precision	Precision Recall F1	Score Training Set	Size Test Set Size	Strategy Type
Remain active for 15 consecutive days	0.71605	0.67308	0.85366	0.75269	20	80	Maximum consecutive active days
Remain active for 20 consecutive days	0.71605	0.67308	0.85366	0.75269	20	80	Maximum consecutive active days
Remain active for 25 consecutive days	0.71605	0.67308	0.85366	0.75269	20	80	Maximum consecutive active days
Remain active for 30 consecutive days	0.71605	0.67308	0.85366	0.75269	20	80	Maximum consecutive active days
Remain active for 35 consecutive days	0.71605	0.67308	0.85366	0.75269	20	80	Maximum consecutive active days
Remain active for 40 consecutive days	0.71605	0.67308	0.85366	0.75269	20	80	Maximum consecutive active days
Remain active for 45 consecutive days	0.71605	0.67308	0.85366	0.75269	20	80	Maximum consecutive active days
Remain active for 50 consecutive days	0.71605	0.67308	0.85366	0.75269	20	80	Maximum consecutive active days
Remain active for 55 consecutive days	0.71605	0.67308	0.85366	0.75269	20	80	Maximum consecutive active days
Remain active for 60 consecutive days	0.71605	0.67308	0.85366	0.75269	20	80	Maximum consecutive active days
Total posts: 10	0.70370	0.66038	0.85366	0.74468	20	80	Total posts
Total posts: 20	0.69136	0.64815	0.85366	0.73684	20	80	Total posts
Total posts: 30	0.69136	0.64815	0.85366	0.73684	20	80	Total posts
Total posts: 40	0.69136	0.64815	0.85366	0.73684	20	80	Total posts
Total posts: 50	0.69136	0.64815	0.85366	0.73684	20	80	Total posts
Total posts: 60	0.69136	0.64815	0.85366	0.73684	20	80	Total posts
Total posts: 70	0.69136	0.64815	0.85366	0.73684	20	80	Total posts
Total posts: 80	0.69136	0.64815	0.85366	0.73684	20	80	Total posts
Total posts: 90	0.69136	0.64815	0.85366	0.73684	20	80	Total posts
Total posts: 100	0.69136	0.64815	0.85366	0.73684	20	80	Total posts

for studies showed that allocation interval did not affect data. The outbreak was still in lockdown, so intercity travel needed a one- to two-week quarantine. A 15-day criterion fails to account for this obligatory isolation period and may misclassify medium-term visitors as transitory tourists. Longer durations better capture persistent sentimental patterns associated with chronic inhabitation. This study chose “active time range of more than 30 days” to better distinguish tourists from locals after careful consideration. After removing international tourist records, 120,319 valid data items were maintained for this study’s emotional database.

Disorders and missing values occurred in the data during the collection, recording, and processing procedures because certain housing data were still being updated. For this study, the housing data from 1 January 1900 to 31 December 2022, were cleaned in the following ways in order to remove the detrimental influence of aberrant housing data on modelling. Initially, samples that were missing a significant amount of fundamental data and information on architectural attributes were removed from the dataset. Secondly, relevant values of samples with comparable architectural attributes were substituted if just a small amount of irrelevant information was absent [38]. Ultimately, Chongqing CUA retained 4865 housing data in total.

After clipping according to the research scope (accessed on 6 June 2024), there were 101,954 points of interest (POIs) located within Chongqing CUA that were obtained.

3. Methods

This study considers RBs and the surrounding built environment part of the community. NLP examined the link between PD and RBE indicators in Chongqing’s central urban area (CUA) using multi-source data and the grid method. We investigated PD predicting factors across RBE levels using 5881 RBE indicator samples from mountainous cities. The following four goals must be met: Using natural language processing and spatial kernel density distribution to locate PD in Chongqing CUA; using a weighted overlay with raster data and RBE classification with the natural breakpoint method to determine how PD is related to environmental factors in space; and using nonlinear regression analysis to examine the relationship between PD and environmental factors. As shown in Figure 2, this research provides the theoretical framework and technical direction for understanding and enhancing PD distribution in RBE during urbanisation.

3.1. PD Analysis Based on NLP and Manual Correction

To identify texts containing perceptual density evaluations of specific locations, this study implemented a systematic text processing pipeline (Figure 3). Following Chinese language rules [39], English, numerical, and other irrelevant terms were then manually filtered out. Python’s Jieba tokenisation system was used to segment comment texts, followed by manual identification of 323 location-specific terms to construct a Chongqing location lexicon. Daniel & Stokols [40] distinguished between “crowding” and “density”, laying a foundational theoretical framework for PD research. He emphasised that the relationship between environmental density and subjective crowding is not linear, highlighting the critical role of psychological perception. Evans & Wener [41] explored the psychological mechanisms of perceived crowding in enclosed spaces, introducing key concepts such as “oppressiveness”, “loss of control”, and “stimulus overload”. Mouratidis & Kostas [42] examined the connection between PD and social emotions, analysing how high-density environments influence well-being and stress. However, there is still a lack of a systematically developed lexicon to describe PD. Guided by Stokols’ definition of PD, we categorised high PD into four dimensions: crowding, oppression, overcrowding, and compactness; and low PD into four corresponding dimensions: uncrowded, loose layout, sparseness, and a sense of openness. Based on this classification, we manually screened relevant keywords to construct a PD lexicon (

Table 2. PD Dictionary Type Description and Keywords.

Perceived Density (PD)	Type Description	Keywords
High PD	Crowding Perception	Packed, Intensity, Layered
	Oppressive Feeling	Stress, Tension or Nervousness, Emo, High Risk, Surreal, Fear, Dizziness, Danger, Depression, Oppression or Repression, Frightening or Scary, Suffocation or Choking, Horror or Terrifying, Forced, Shock or Impact, Insignificant
	Overcrowded	People Coming And Going, Crowded, Foot Traffic, Sea of People, Overcrowded
	Compactness	A Lot of, Multiple, Dense, Crammed or Tightly Packed
Low PD	Uncrowded	Relaxed
	Loose Layout	Lighthearted or At Ease, Composed, Calm or Relaxing
	Sparseness	Increasingly Sparse, Desolate or Spacious, Empty
	Sense of Openness	Open Space, Vast, Expansive, View or Visual Field or Wide View

).

Using the Chongqing location lexicon and NLP technologies, we initially filtered 37,199 location-tagged comments from the dataset. Subsequent application of the PD lexicon identified 1363 texts containing density evaluations. To address potential semantic ambiguities in Chinese linguistic contexts, these results underwent manual verification, yielding a final dataset of 1125 validated comments for subsequent sentiment analysis of PD and its environmental impacts in high-density residential areas.

3.2. Sentiments Analysis of PD Based on NLP and Sentiment Lexicon

This study utilised the NLP technology of the Python 3.9 to analyse the sentiments in Weibo comments. Known for its ease of use and high accuracy, it is ideal for processing large volumes of text and is widely used by non-experts [43,44].

In the sentiment analysis phase, this study adopts the Dalian University of Technology’s Sentiment Lexicon (DUT Emotion Lexicon) as the base dictionary. This lexicon categorises emotions into seven primary types: “joy”, “good”, “anger”, “sorrow”, “fear”, “disgust”, and “surprise”), along with 21 subcategories.

In this classification framework, “joy,” “good,” and “surprise” are grouped as positive sentiments, while “anger,” “sorrow”, “fear” and “disgust” are grouped as negative sentiments.

To facilitate computation, the traditional sentiment polarity classification (neutral = 0, positive = 1, negative = 2) is modified by assigning a value of −1 to negative sentiments, thereby offering a more intuitive numerical representation of sentiment polarity (

Table 3. Part of negative and positive vocabulary displays.

Weight	Part of the Vocabulary
−1	no, not, can’t, not much, don’t have to, didn’t, no, don’t, none, non, don’t, in vain, empty, in vain, in no way…
1	yes, affirmative, positive, can, have, do have, some, available, exist, do, have, fruitful, successfully…

). Degree adverbs were assigned weight values according to a gradient drop formula [45] (

Table 4. Part of the degree adjective vocabulary displays.

Degree	Weight	Part of the Adverbs of Degree	Number
1	3	very, extremely, fully, absolutely, most	69
2	2.1	super, over, excessive, more than, bias, extra	30
3	1.5	quite a lot, especially, extraordinarily, greatly	42
4	1.06	more, more and more, also, further	37
5	0.75	slightly, a little, somewhat	29
6	0.53	not a little, not very, not much, relatively	12

).

$$\begin{array}{c}{T}_{k+1}=T_1{\left({\displaystyle \frac{\sqrt{2}}{2}}\right)}^k\end{array}$$

(1)

where T₁ is the weight value of the first level “extremely, most”; the constant ${\displaystyle \frac{\sqrt{2}}{2}}$ is the gradient descent rate. The grading of the degree adverbs used in this study is shown in Table 4.

To more fairly represent the spatial distribution of sentiments in the study region and to further explore their numerical distribution characteristics instead of focussing on individual points, this study employed a 1 × 1 km grid to analyse the PD, combining the total area of unit sentiments with both positive and negative sentimental scores, as illustrated in Figure 4.

3.3. Indicator System of Environmental Factors

3.3.1. Selecting Environmental Factors

The indicator system considers mountainous cities’ natural and constructed environments, including topography and watersheds. The indicator system includes precipitation, temperature, and humidity indications for human thermal comfort. Built environment includes transport and green spaces. Physical density covers BD and FAR, while socio-cultural encompasses everyday life services and cultural and educational institutions (

Table 5. Indicators affecting the PD of residents in high-density RBE.

Types of Influencing Factors	Level I Indicators	Level 2 Indicators	Unit
Natural and built environment factors	Natural geography environment	Terrain altitude	m/km2
		Water area	km2
		Annual average precipitation, PRE	mm
		Annual average relative humidity, RHU	%
		Annual average temperature, TEM	°C
	Road and traffic	Road density	km/km2
		Bus stops	point/km2
		Subway stations	point/km2
	Green and open spaces	Park squares	point/km2
		Scenic spots	point/km2
Physical density factors	Building density, BD	Average value of residential BD in unit grid	null
	Floor area ratio, FAR	Average value of residential FAR in unit grid	null
Socio-cultural factors	Daily life service facilities	Restaurants	point/km2
		Leisure and recreation	point/km2
		Shopping services	point/km2
		Medicine services	point/km2
		Life services	point/km2
		Fitness sports	point/km2
	Cultural and educational institutions	Educational institutions	point/km2
		Cultural heritage	point/km2

).

3.3.2. Avoid Multicollinearity

As Figure 5 illustrated, for each group of attributes, we retained as many available variables as possible while conducting variance inflation factor (VIF) tests and Pearson correlation analysis to study the multicollinearity problem. Variables with high collinearity (VIF > 10) were removed (Figure A1). It is worth noting that when two variables exhibit high collinearity, we retain the one that contains more comprehensive information or is in line with the interests of this study. For example, TEM, PRE, and RHU showed strong collinearity (Figure A2). We retained the former because the temperature caused by the mountainous thermal environment effect in Chongqing directly determines the feasibility and comfort of outdoor activities, thus significantly affecting people’s perception of spatial density.

3.3.3. Model Architecture

We fit all selected independent variables to the OLS model with each data level to determine their explanatory power and generate a baseline model [46]. To find spatial dependence, Moran’s I test was performed on OLS regression residuals.

Notably, given the substantial differences in scale among the variables, a logarithmic transformation was applied to both sides of the regression equations, resulting in a log-log model. This specification allows the coefficients to be interpreted as elasticities—that is, the percentage change in negative sentiments corresponding to a one percent change in each independent variable [47,48]. In practical terms, each coefficient represents the estimated percentage change in positive sentiment value associated with a one percent change in the explanatory variable.

4. Results

4.1. Distributions of Residents’ PD and Sentiments in the Study Area

High PD areas (Figure 6a) were mainly concentrated in Yuzhong District and surrounding core urban areas such as Jiulongpo, Shapingba, Nan’an, and Jiangbei, forming a clear “central clustering” pattern. These hotspots (in red and orange) showed strong spatial aggregation, likely influenced by high population density, commercial activity, and the presence of landmark buildings. In contrast, low PD areas (Figure 6b), though also present in central districts, were less dense, with fewer and more scattered red zones, indicating a dispersed, non-central distribution. Figure 6c indicated a clear gradient in emotional structure under high PD. As the context shifted from “compactness” to “crowding–oppressiveness–overcrowding”, positive sentiments—particularly good ones—gradually declined, while negative sentiments increased. Among them, disgust and fear consistently dominated, sadness rose significantly in oppressive settings, and anger peaked under overcrowding. Overall, high-density contexts exhibited a negative pattern characterised by a “disgust/fear-dominated” structure. In contrast, low PD presented a stable “positive–low negative” configuration, with happy and positive states taking clear precedence and negative sentiments remaining minimal (Figure 6d). Only under “sparseness” does sadness show a slight increase, reflecting a sense of emptiness, but this did not alter the predominance of positive sentiments. Notably, “sense of openness” and “uncrowded” conditions yielded the highest proportions of positive sentiments.

4.2. Spatial and Quantitative Distribution Characteristics of Residents’ PD and Sentiments

Table 6. The quantitative distribution of PD in CUA of Chongqing.

Perceived Density (PD)	Number	Proportion	Average Positive Sentiment Value	Average Negative Sentiment Value
High PD	664	71.32%	4.436747	1.986446
Low PD	267	28.68%	6.985019	1.629213

shows that areas with high PD received more public comments but had lower average positive sentiment and higher negative sentiment compared to low-density areas. Figure 7a–d revealed that spatially, positive sentiments in high-density areas were concentrated within urban cores, whereas negative sentiments exhibited a more diverse spatial distribution pattern.

In contrast, low-density areas displayed a more widespread distribution of faintly negative sentiments. In terms of emotional quantity distribution, areas with high PD exhibited significantly higher positive sentiment counts than those with low PD, with “sense of compactness” and “sense of oppression” emerging as the most prominent dominant emotions.

In view of the volume of research, the following section focused on the positive sentiments associated with PD as the research subject, exploring the environmental factors that influenced their variation.

4.3. Spatial Correlation Between Residents’ Sentiments of PD and Physical Density Factors

4.3.1. Sentimental Univariate Local Spatial Autocorrelation of PD

The findings, analysed using the Geoda platform, revealed distinct spatial patterns in sentimental responses associated with PD (Figure 8). In high-density environments, positive sentiments tended to cluster and spill over into adjacent areas. “High–high” clustering hotspots and “high–low” transitional zones indicated strong spatial spillover effects. Meanwhile, negative sentiments indicated localised clusters. A “low–low” cold spot and “low–high” anomaly pattern suggested localised and discrete distribution.

4.3.2. Bivariate Spatial Autocorrelation Analysis Between Sentiment of PD and Physical Density

Physical density is a critical determinant of public health. Given that positive and negative sentiments were inversely related, and considering this study’s focus on improving public health in high-density residential environments, the analysis emphasised positive sentiments under conditions of high PD.

The bivariate local spatial autocorrelation between positive sentiments and BD in high-PD areas (Figure 9a–c) yielded a global Moran’s I value of −0.039, indicating an overall weak and statistically non-significant spatial association. Nonetheless, local significance testing identified a limited number of spatial clusters. A “low–high” pattern (high BD accompanied by low positive sentiments) emerged sporadically across the study area, while the central urban core exhibited both “high–high” and “high–low” clusters, suggesting that zones of high BD simultaneously contained areas with either elevated or diminished levels of positive sentiments.

Similarly, the bivariate local spatial autocorrelation between positive sentiments and FAR (Figure 9d–f) produced a Moran’s I of −0.040, again reflecting an overall weak and non-significant global correlation. At the local level, scattered “low–high” clusters (high FAR associated with low positive sentiments) were observed, while the central districts exhibited both “high–high” and “high–low” patterns. These findings imply that although the general relationship between development intensity and positive sentiments is weak, localised spatial heterogeneity persists, with high-intensity built environments associated with varying emotional outcomes depending on specific geographic and socio-environmental contexts.

4.4. The Construction of the Five-Level RBE

The above research results indicated that positive sentimental experiences in high PD areas differed between core and surrounding edge areas. While, overall, physical density was negatively correlated with positive sentiments in high PD areas, spatially, areas with high BD and high floor area ratio in the core zones tended to accumulate positive sentiments, whereas low physical density in the peripheral areas was more likely to trigger positive sentiments. In contrast, physical density was positively correlated with negative sentiments in low PD areas, meaning that an increase in physical density led to an increase in negative sentiments. However, spatially, high BD and high floor area ratio in the core areas tended to trigger clusters of negative sentiments in low PD areas, whereas low physical density in the peripheral areas was associated with lower levels of negative sentiments. Therefore, the study introduced a tiered mechanism to spatially and statistically analyse the factors influencing positive sentiments in high PD areas across different physical density zones in high-density residential environments.

According to the SPSSAU (https://spssau.com/, (accessed on 25 July 2024)) results, FAR accounted for 39.19% of the weight (entropy = 0.7483; utility = 0.2517), while BD contributed 60.81% (entropy = 0.6093; utility = 0.3907). Standard techniques for assessing residential density are equal-interval, natural breakpoint, and interquartile methods. The equal-interval and quantile approaches partition data into predetermined ranges or groups, disregarding the original data distribution and potentially overlooking critical information. The natural breakpoint method, which maximises within- and between-class variance, is more effective for differentiating similar values [49]. This study employed the natural breakpoint method to categorise residential density. We standardised BD and FAR in ArcGIS, assigned specified weights, and categorised them into five levels of RBE (Figure 10).

RBE 1 (958 buildings, 5291 grids) accounted for 19.70%. The minimum value for building indicators in this region was 0, the maximum value was 5, the mean score was 1.053, and the standard deviation was 0.284.

RBE 2 (1915 buildings, 348 grids) accounted for 39.36%. The minimum value for this region was 1, the maximum value was 5, the mean score was 1.754, and the standard deviation was 0.691.

RBE 3 (1086 buildings, 140 grids) accounted for 22.32%. The minimum value for this region was 1, the maximum value was 4, the mean score was 2.070, and the standard deviation was 0.690.

RBE 4 (777 buildings, 82 grids) accounted for 15.97%. The minimum value for this region was 1, the maximum value was 5, the mean score was 2.177, and the standard deviation was 0.731.

RBE 5 (129 buildings, 20 grids) accounted for 2.65%. The minimum value for this region was 1, the maximum value was 4, the mean score was 2.181, and the standard deviation was 0.833.

Positive sentiments in high PD areas increased sharply from RBE2 to RBE3 and from RBE4 to RBE5, with RBE5 and RBE3 showing the highest levels of positive sentiment.

4.5. Proportion Characteristics of Environmental Factors in the Five-Level RBE

Regarding natural geography, consistent with the notion of choosing flat terrains for urban development, altitude diminished as PD increased. Road density increased in high-density regions, while the concentration of bus stops and subway stations was highest in medium-density and low-density locations. RBE 3 possessed the most extensive water area, whereas RBE 5 exhibited the least (Figure 11). In terms of physical density, Figure 12 showed that RBE 5 had the highest BD at 44% and the lowest FAR at 2.906. RBE 1 had the lowest values (BD = 14%, FAR = 1.573). Services exhibit an unequal distribution across various residential areas in relation to socio-cultural density. The socio-cultural density in low-density areas is inferior to that in high-density ones (Figure 13). RBE1 comprises merely 3.59% of leisure and recreation and 2.92% of restaurants. On the other hand, RBE4, located in the high-density region, has the majority of resources, especially when it comes to leisure and recreation (72.24%) and medical services (40.00%).

4.6. Associations Between Positive Sentiments of High PD and Environmental Factors in the Five-Level RBE

4.6.1. The Heatmap of Pearson Correlation Analysis

The correlation heatmap findings revealed substantial disparities among the components in various RBEs (Figure 14). In general, the correlation between environmental density and socio-cultural factors in the low-value group (RBE 1–3) was weak. This means that these factors did not do a good job of explaining the outcome variables at lower levels. In the medium-high value group (RBE 4), the positive correlation between socio-cultural factors (including catering, shopping, medical care, education, and cultural heritage) and the outcome variables was most pronounced (correlation coefficient of approximately 0.30–0.39). Additionally, certain traffic factors (such as buses, subways, squares, and parks) exhibited a significant positive impact. When entering the high-value group (RBE 5), these action patterns changed, though. There was a strong negative correlation between social and cultural factors like medical care, living services, and education and the outcome variables (−0.46, −0.37, −0.29). Physical density factors and some traffic factors also showed a negative relationship.

4.6.2. OLS Regression Analysis

Table 7. The OLS result for the CUA of Chongqing.

Log-Log, Sample Size 5881
Model		OLS
R2		0.206
Adj. R2		0.203
F-statistic		79.9
AIC		942.8
Moran’s I on residual		−0.0153
p_value		0.02281
Evalution Index		VIF	coef	p > |t|
RBE value		2.058	0.0366	***
Natural and built environment factors	Terrain altitude	1.082	0.0067	*
	Water area	1.032	0.0061	*
	TEM	1.138	0.0017
	Road density	1.666	0.0063
	Bus stops	1.387	−0.0111	**
	Subway stations	1.308	0.0168	***
	Park squares	1.523	0.0086	**
	Scenic spots	2.443	0.0178	***
Physical density factors	BD	1.447	−0.0063
	FAR	1.539	−0.0181	***
Socio-cultural factors	Restaurants	2.799	0.0396	***
	Leisure and recreation	3.796	0.0502	***
	Shopping services	4.308	−0.0376	***
	Medicine services	3.822	−0.008
	Life services	5.508	0.0043
	Fitness sports	2.462	0.0325	***
	Educational institutions	3.501	0.0042
	Cultural heritage	1.036	0.0109	**

Notes: while the significant levels of p > |t| at 0.001, 0.05 and 0.1 are shown as ***, **, *, The selected variables with VIF < 10.

provides the OLS results for the CUA of Chongqing. This section mainly provides the overview of model fit and significant variables. The interpretation and relevance to prior studies will be discussed in the following section.

The model fit (R² = 0.206, Adj. R² = 0.203) indicates that natural and built environment features, material density, and socio-cultural factors collectively explain approximately 20% of the variation in positive sentiments. More importantly, the residuals show a significantly negative Moran’s I value (−0.0153, p = 0.0228), suggesting weak spatial dispersion in the errors rather than strong autocorrelation. The low magnitude of spatial autocorrelation supports the reliability of the parameter estimates.

Several factors—including value, terrain altitude, water area, bus stops, subway stations, park squares, scenic spots, FAR, restaurants, leisure and recreation facilities, Shopping services, fitness and sports venues, and cultural heritage sites—were identified as significant predictors. In contrast to previous studies, temperature did not exhibit a statistically significant effect. This may be attributed to the social restrictions during the lockdown period, which could have diminished people’s perception of microclimatic conditions, thereby allowing social and built environment factors to outweigh meteorological influences.

5. Discussion

Although previous studies have shown that high-density living environments may have a positive psychological impact on the public, this study clarifies that in mountainous cities, residents living in high-density environments are positively affected by the surrounding natural, built, and socio-cultural environmental factors.

5.1. Residents’ PD in Hierarchical RBEs from an Affective Dimension

Our research indicates that areas with higher PD elicit a greater quantity of positive emotions; however, these emotions exhibit reduced diversity and intensity. This suggests that under high-PD conditions, positive emotional experiences may be qualitatively inferior, which diminishes their perceptual salience and impedes the cognitive reflection necessary for their consolidation into enduring sentiments. This apparent degradation in emotional quality aligns with broader adaptive mechanisms observed in dense urban settings. For instance, Mouratidis [50] found that prolonged residence in high-density areas can enhance emotional resilience and reduce sensitivity to shared spaces. Nevertheless, such adaptive capacity appears to have limits. While environmental and socio-cultural factors exert minimal emotional influence in low-PD regions, medium- to high-PD areas benefit markedly from integrated socio-cultural amenities and transportation infrastructure. Conversely, in the highest-density category, this beneficial correlation reverses—a pattern corroborated by Kan et al. [14] in studies of high-density residential zones in Hong Kong. Their work further revealed that extreme crowding and inadequate personal space can compromise privacy, elevate stress, intensify interpersonal conflict, and ultimately disrupt emotional adaptation. To systematically investigate these nonlinear dynamics within Chongqing’s unique urban context, this study employed two fundamental morphological indicators—FAR and BD—alongside the natural breaks method to classify the central urban area into five graded levels of built environment density. This classification framework captures the city’s complex three-dimensional structure more effectively than conventional metrics such as average building height or BD alone and differs from clustering analyses based on dynamic population distributions [51] or composite typologies integrating building height and community function [52]. In contrast to international and domestic analogues—such as terrain-adjusted density models applied in coastal mountainous cities like Qingdao [53] or spatial syntax approaches used in studies of cities like Busan [54]—this study proposes a density classification system tailored to high-intensity mountainous urban areas. Our framework integrates morphological and intensity-based indicators with concrete physical and statistical significance and emphasises the aggregate effects of socio-cultural and transportation facilities. It also helps elucidate the nonlinear turning points within density–emotion relationships. By adopting multi-dimensional and heterogeneous spatial grouping, this approach clarifies the complex mechanisms underlying density perception and emotional response, consistent with the advanced spatial modelling techniques applied in other mountainous urban areas worldwide. The classification framework, indicator selection, and mechanistic interpretation presented here demonstrate both uniqueness and regional adaptability—particularly in identifying behavioural-emotional reversal phenomena within ultrahigh-density groups—thus offering new theoretical and practical insights for the field.

5.2. Influence of Environment Factors in RBEs of Mountainous Cities

5.2.1. Natural and Built Environmental Attributes

Fundamental natural environments—terrain and water—exert a complex and nuanced influence on the constructed environment. At a 10% significance level, terrain altitude and water area exhibited weak positive effects. Elevated elevations may offer an expansive perspective and enhanced ventilation [55], which is implicitly linked to reduced health risks and diminished despair during the lockdown, highlighting the importance of air circulation and social separation. The observable expanse of water bodies, such as rivers and lakes, constitutes a limited natural landscape resource that can alleviate anxiety through two mechanisms: its aesthetic appeal calms the emotions [56,57]; and the waterfront environment typically offers greater openness, contrasting with the adverse connotation of “crowdedness” and symbolising the “breathability” of both physical and psychological space. Despite the minimal effect size, this implies that natural geographical features may function as an emotional backdrop. The enhancement of transportation and recreational infrastructure elevated the constructed environment. The accessibility of tube stations (coefficient = 0.0168, p < 0.001) significantly enhanced positive opinions. This data corroborates our hypothesis that throughout the lockdown, the public transport system, which sustained the city, functioned not merely as a mode of transit but also as an emblem of urban resilience, social connectivity, and life security, alleviating feelings of isolation and worry. Bus stops, an alternative means of transportation, are fundamentally different. Traffic, weather, and other factors frequently delay buses, resulting in congestion and affecting residents’ attitudes. The public perception is that buses are less convenient than subways [58]. Furthermore, park squares (coef = 0.0086, p < 0.01) and scenic places (coef = 0.0178, p < 0.001) have considerable positive effects. This indicates that visible and accessible green and recreational spaces, whether in person or through social media, modulate moods even when activities are limited and are essential urban elements for psychological well-being [59,60].

5.2.2. Physical Density’s Attributes

High physical density presents a dual challenge for metropolitan inhabitants. The FAR was negatively correlated with positive sentiments (coefficient = −0.0181, p < 0.001). Rapid urban expansion and public demand create high-density constructed settings, yet overdevelopment can make lockdowns more uncomfortable for local inhabitants. Many think high-density buildings are crowded, dark, and unventilated. Such surroundings are associated with pandemic risk, which may worsen emotional responses [61].

5.2.3. Socio-Cultural Attribute

Socio-cultural factors cause the most complex consequences. Recreation, exercise, and restaurants had the highest positive coefficients. The main places city residents live, shop, and socialise are these. A “visible convenience” might provide people a sense of routine and hope, even if they cannot be used in person. In contrast, shopping services showed a significant negative correlation (coef = −0.0376, p < 0.001). These facilities may be associated with lockdown stressors such waiting in lines, panic buying, and running out of supplies, which could generate negative sentiments [62]. The positive impact of cultural heritage (coef = 0.0109, p < 0.01) may stem from its role in fostering local identity and communal memory, providing psychological stability during turbulent times [63].

5.3. Strategies for Urban Managers and Designers in High-Density Mountainous Cities

The interplay of natural and built environments, physical density, and socio-cultural factors jointly promotes PD. Urban managers must advocate for suburban parks in low-density areas in appropriate, low-lying zones and develop measures for green space coverage and accessibility to guarantee that natural resources serve all citizens. In terms of physical density, the Chongqing government has implemented building regulatory laws for the central districts, taking into account the previously mentioned threshold issue. The Chongqing Detailed Planning Guide (Figure 15) shows that RBE1–2 mostly live in high-intensity zones, which is in line with FAR and BD rules. On the other hand, RBE3–5 needs stricter BD rules (

Table 8. Residential building indicators regulations in Chongqing Detailed Planning Compilation Guide (https://ghzrzyj.cq.gov.cn/ (accessed on 21 July 2024)).

Land Use	Type	Zone 1	Zone 2	Zone 3	Zone 4	Zone 5
Residential area	Base FAR	1.2	1.5	2	2.5	2.5
	Upper limit of FAR	1.8	2.3	2.5	3	3
	Upper limit of BD	≤40%	≤35%	≤35%	≤30%	≤30%

). The government ought to implement more stringent controls to curtail the swift expansion of urban residential zones.

Based on the quantitative relationships identified in this study between urban environmental factors and residents’ emotional responses, and considering the spatial characteristics of Chongqing as a high-density mountainous city, the following planning recommendations are proposed:

5.3.1. Prioritise High-Impact Amenities with Differentiated Allocation

Leisure and recreation facilities (a 1% increase is associated with a 0.0502% increase in positive sentiments among residents in high PD areas), restaurants (a 1% increase corresponds to a 0.0396% increase), and fitness facilities (a 1% increase corresponds to a 0.0325% increase) exhibit the strongest positive effects. Therefore, these amenities should be prioritised in medium- to high-density residential areas (RBE 3–4). Considering spatial constraints, they may be accommodated through the adaptive use of stairway platforms, elevated walkways, or underutilised slopes. Figure 16a depicts fitness equipment installed on a slope in Chongqing.

Cultural heritage sites (a 1% increase corresponds to a 0.0109% increase in positive sentiments) also contribute positively, though with a relatively modest effect. Low-intensity and adaptive reuse strategies—such as interpretive signage and seating areas—are recommended to enhance cultural identity and emotional engagement without extensive construction. The famous Yangtze River Cableway in Chongqing, which was previously a daily way for locals to traverse the river, is seen in Figure 16b. It is now a well-known landmark that serves to both preserve the city’s collective memory and spread awareness of this mountain city’s culture.

In contrast, shopping services (a 1% increase corresponds to a 0.0376% decrease in positive sentiments) and bus stops (a 1% increase corresponds to a 0.0111% decrease) may induce emotional fatigue under conditions of high density. Therefore, large-scale commercial complexes should be restricted in high-density areas, with planning priorities given to small-scale, decentralised convenience retail formats. Bus stops should be improved through the provision of shading, seating, and real-time information systems to mitigate the negative effects of overcrowding.

5.3.2. Implement Density-Adaptive Emotional Buffering

FAR (a 1% increase corresponds to a 0.0181% decrease in positive sentiments) demonstrates a significant negative association with residents’ sentiment. When FAR exceeds 5.0, development projects should be required to incorporate public open spaces, such as pocket parks, rooftop gardens, or ventilated corridors, to alleviate psychological pressure associated with high density. In the highest-density zones (RBE 5), small-scale emotional buffering interventions are recommended, including resting platforms along stair climbs, view corridors orientated towards green or water features, and decentralised service kiosks to reduce crowding effects. Figure 16c depicts a pocket park within a community in Yuzhong District, Chongqing.

5.3.3. Enhance Synergies Between Transit, Nature, and Sentiment

A 1% rise in tube stations and beautiful locations enhances positive sentiments by 0.0168% and 0.0178%, respectively. Clear signage, sheltered pathways, and elevated walkways where topographical constraints necessitate should enhance pedestrian access to adjacent landscapes and green spaces within a 500 m service radius of metro stations. Li Zi Ba, seen in Figure 16d, represents the metro train runs right through the middle of a building in Chongqing. The exceptional design of a train traversing the structure enhances the visual impact of high-density development and fosters a more favourable opinion of such building. A 1% elevation in terrain altitude (resulting in a 0.0067% rise in good feelings) and water bodies (yielding a 0.0061% increase) each exert moderate influences; nevertheless, their combined impacts significantly enhance emotional repair and environmental liking. Urban design should stress the preservation and utilisation of natural assets, such as sloped paths and coastal promenades, to enhance inhabitants’ access to nature and mental well-being.

5.4. Limitations

We acknowledge several limitations associated with the use of social media data. First, the varying popularity of Weibo across demographic groups may affect the representativeness of the sample [64,65]. Additionally, users are more inclined to post positive content, introducing potential sentiment bias in geotagged posts [66].

The self-constructed perceptual sentiment dictionary also involves a degree of subjectivity, particularly in term selection, weight assignment, and contextual interpretation. Although the dictionary builds upon an existing foundational lexicon and employs contextual validation to reduce bias, its construction remains dependent on manually defined rules and annotation strategies. While the added terms improve polarity prediction [67], they may not fully capture contextual nuances in emotional expression [68]. For example, some sentiment words related to density perception may exhibit opposite polarities across different urban or cultural settings [69]. To improve the robustness and generalisability of the dictionary, future research could incorporate automated sentiment construction methods based on machine learning or deep learning [70] and combine these with field survey data—such as questionnaires or interviews—for cross-validation [71]. Cross-regional and cross-cultural comparative studies could also help identify and correct systematic biases [72].

Although the low spatial autocorrelation in the residuals suggests that the assumption of independent errors is not severely violated, this does not necessarily indicate that all spatial effects have been captured or that model specification bias is eliminated. Comparisons with spatial regression models (e.g., SAR, SLM, SEM, or GWR) would provide a more rigorous basis for evaluating whether spatial dependence has been adequately addressed [48].

Finally, the temporal scope of this study is limited to public posts from Chongqing in 2022. Future research could include comparisons with other time periods to enhance the generalisability of the findings.

6. Conclusions

This study examines the role of PD in moderating the relationship between urban environmental elements and residents’ sentiments in a high-density mountainous city. The OLS regression results indicated that environmental variables only explained 20.3% of the variance in positive sentiments within high-perceived-density areas. This illustrates that environmental effects on sentimental responses are highly context-dependent and exhibit nonlinear dynamics, demonstrating clear threshold effects under varying density conditions.

First, socio-cultural facilities—including restaurants, shopping, fitness, and cultural heritage—exerted significant positive effects on emotions in medium-to-high density areas, where public transportation and park accessibility provided essential support. This aligns with existing literature emphasising the psychological benefits of urban amenities in moderately dense environments. However, in the highest-density zone (RBE 5), the sentimental effects of these facilities reversed, indicating a density threshold beyond which service saturation may trigger emotional adaptation or overload. This finding challenges conventional assumptions about linear dose–response relationships in environmental psychology.

Second, the methodological integration of perceptual quantification with spatial-statistical modelling offers a replicable framework for emotion-sensitive analysis in topographically complex cities. Despite this contribution, the generalisability of the identified density thresholds may be limited by the singular mountainous context. Future research should validate these relationships across diverse geographic and cultural settings through comparative studies and longitudinal designs.

Finally, these quantified insights carry concrete implications for public policy: rather than uniformly increasing service provision, planners should adopt density-aware facility allocation strategies. In regions of RBE 5, policy should prioritise emotional buffer zones—such as pocket parks and decentralised service nodes—to mitigate perceptual overload. During public health crises, emergency plans should account for elevated emotional vulnerability in high-density areas through targeted mental health support and mobile service delivery.

In summary, this research moves beyond descriptive findings to establish empirically grounded density thresholds and statistical evidence of emotional reversal effects. It provides a scalable methodology for context-sensitive urban planning and emphasises the need for precision interventions calibrated to local density conditions.