Uncovering Drivers of Resident Satisfaction in Urban Renewal: Contextual Perception Mining of Old Community Regeneration Through Large Language Models

Abstract

Urban regeneration has increasingly become a global strategy for promoting sustainable urban development, with the renewal of deteriorating residential communities serving as a key dimension of this process. Within the framework of a people-centered development paradigm, growing attention has been directed toward the necessity of securing residents’ satisfaction in community renewal initiatives. This study employs advanced textual analysis of resident submissions collected from government–citizen interaction platforms to investigate the determinants of satisfaction with renewal projects. Leveraging the semantic comprehension capabilities of large language models (LLMs), we identify both salient keywords and sentiment orientations embedded in residents’ narratives. Guided by the theoretical framework of resident satisfaction, the extracted keywords are organized into seven thematic domains: basic infrastructure improvement, quality-enhancement renovation, solicitation of residents’ preferences, residents’ decision-making power, policy transparency, construction governance, and community-level communication. Regression modeling is subsequently applied to assess the relative influence of these thematic domains on residents’ satisfaction. The findings suggest that insufficient integration of residents’ preferences at the preliminary stages of participation constitutes a principal source of dissatisfaction during the implementation of renewal projects. Furthermore, the study compares Latent Dirichlet Allocation (LDA) topic modeling with LLMs-based topic clustering, revealing the latter’s superior capacity to capture thematic structures in complex, long-form textual data. These results underscore the potential of LLMs to enhance the analytical rigor of research on urban regeneration and citizen participation.

1. Introduction

Community renewal has become a key livelihood-oriented component of urban regeneration strategies across the globe, playing a significant role in improving public living standards and promoting sustainable economic development worldwide [1,2]. The 2030 Agenda for Sustainable Development, adopted at the United Nations Sustainable Development Summit, emphasizes the goal of “making cities and human settlements inclusive, safe, resilient, and sustainable” [3]. This agenda has encouraged community renewal as an essential approach, including the upgrading of slums and informal settlements. By providing high-quality housing and public spaces, it advocates integrated and participatory strategies that mitigate socio-economic segregation and gentrification [4,5].

As both a participant in and active promoter of the United Nations Sustainable Development Goals (SDGs), China has incorporated sustainable development as a fundamental national policy, embedding it within the 14th Five-Year Plan and the long-term vision for 2035 [6]. At present, China faces dual pressures: a vast stock of aging residential communities and an accelerating demographic shift toward an aging population. To address these challenges, old community renewal—particularly the upgrading of age-friendly facilities and elderly care services—has become essential to improving urban livability [7]. However, in practice, renewal projects often encounter multiple challenges, including irrational infrastructure planning and design, conflicts between individual and collective interests of residents, difficulties in resolving long-standing historical issues, weak awareness of resident participation in construction and management, and insufficient mechanisms for multi-stakeholder negotiation. These issues have led to generally low levels of resident satisfaction [7,8,9,10]. Such challenges are not unique to China. As Hackney noted, in the large-scale urban regeneration of twentieth-century Britain, the excessive reliance on top-down modernist planning and spatial restructuring not only fragmented community networks but also failed to deliver housing environments that met residents’ expectations, ultimately leading to the gradual erosion of neighborhood bonds and trust in planners [11].

To address the inadequacy of resident satisfaction in current old community renewal initiatives, it is necessary to conduct an in-depth analysis of the challenges facing old community regeneration, identify the determinants of dissatisfaction, and propose scientifically grounded and practical optimization strategies. Given that multiple stakeholders are involved in the renewal of old residential communities in China—with residents serving as the most direct stakeholders [12]—this study adopts a resident-centered perspective. Such an approach not only reflects residents’ genuine needs and expectations regarding community renewal but also enhances their enthusiasm for participation [13]. In turn, this can improve the efficiency and effectiveness of renewal projects, while providing strong support for evidence-based policy-making.

With the development of “Internet Plus Governance,” the People’s Daily Online “Leadership Message Board” has become an important digital platform for interaction between the government and residents [14]. In this context, the direct analysis of residents’ messages on various platforms is critically significant for comprehending public demands. Currently, when existing scholars engage in the analysis of short text corpora, such as unstructured online messages, traditional methodologies, including word frequency-based clustering and topic models (such as latent Dirichlet allocation), frequently exhibit inadequate performance due to text sparsity and insufficient co-occurrence information. The elimination of low-frequency terms during data preprocessing results in the loss of essential semantics, and the manual labeling of topic clusters based on clustering outcomes tends to be subjective [15,16,17]. Concurrently, in recent years, numerous scholars have investigated the application of large language models (LLMs) to examine residents’ experiences and perceptions of fairness in urban renewal from both theoretical and methodological perspectives. Such explorations encompass the simulation of public participation, the elucidation of residents’ needs, the enhancement of human–machine collaboration, and the promotion of transparency in explanations [18,19,20]. In light of this, the present study proposes an innovative analytical framework addressing the challenges associated with the renewal of aging communities, leveraging LLMs in conjunction with resident satisfaction theory, thereby harnessing the natural language recognition and generation capabilities of these models. This approach effectively addresses the limitations inherent in traditional topic models and investigates the application of LLMs in the old community.

2. Literature Review

2.1. Resident Satisfaction in Old Community Regeneration

As one of the key stakeholders, residents play a vital role in the process of old community regeneration [21]. Resident satisfaction, in this context, can be understood as a subjective perception derived from the integration of social satisfaction theory and Maslow’s hierarchy of needs theory. It reflects the extent to which individuals’ needs concerning both themselves and their living environment are fulfilled by society [22].

Currently, questionnaires and interviews constitute the predominant research methods employed in the analysis of resident satisfaction, as they facilitate the direct and detailed collection of specific opinions from residents. However, these methods exhibit inherent limitations. Questionnaires elicit residents’ opinions through predetermined options and cross-sectional temporal frameworks [1,23,24], which may constrain residents from articulating their genuine and nuanced perspectives, thereby inadequately addressing the underlying factors that influence their satisfaction. Concurrently, the limitations associated with interviews include a restricted sample size and their dependence on the subjective inquiries and analyses conducted by the researcher [23,25], which may introduce response bias from residents, thus complicating the generalizability of research findings. As illustrated in

Table 1. Methods of data collection.

Author (Year)	Methodological Details	Methodology
Huang (2017) [23]	Designed a questionnaire scale based on policy regulations and employed probability proportional to size (PPS) sampling by employment type. Over the course of one year, 1000 questionnaires were distributed to three categories of migrant workers, yielding 739 valid responses. Additionally, 32 in-depth interviews were conducted to complement the quantitative analysis.	Survey Questionnaire + Interviews
Pan (2023) [25]	Conducted 15 in-depth interviews with residents, selected through snowball sampling, using a pre-designed interview guide.	Interviews
Chaulagain (2024) [24]	Developed a survey scale based on existing literature and administered a self-completion questionnaire via mobile devices. Screening questions were included, and over the span of one month, 243 valid responses were collected.	Survey Questionnaire
Huang (2025) [1]	Applied stratified random sampling within a specific community, with trained surveyors assisting in one-time distribution and collection of questionnaires, resulting in 676 valid responses.	Survey Questionnaire

.

With the growing prevalence of e-government platforms and social media, researchers are increasingly able to capture residents’ opinions and emotional expressions in a more immediate and precise manner compared to traditional surveys and focus groups [26]. Dorostkar et al. argued that measuring urban sentiment via social media can contribute to advancing sustainable urban development [27]. Likewise, Kontokosta et al., using structured data from the U.S. 311 citizen hotline obtained via the DataKC open platform, conducted spatial aggregation analysis and revealed that service distribution was inequitable in low-income and minority communities due to sociocultural differences, resulting in generally low levels of resident satisfaction [28]. He Ju et al. used natural language processing and spatial analysis on data from the Beijing 12345 hotline. They found that in old community renovations, the government prioritizes basic needs but delays requests for improvements [29]. Zhao et al. utilized Weibo check-in data to capture residents’ spatial activity patterns, which, when integrated with mobile signaling and transportation trajectory data, served as key indicators for assessing community vitality and the effectiveness of urban renewal [30].

2.2. Large Language Models

Large language models (LLMs) are advanced artificial intelligence models built on natural language processing (NLP). They demonstrate remarkable capabilities in both language understanding and generation [31]. However, their performance may vary depending on the complexity of the task, the quality of training data, and the formulation of human prompts [32]. Such models enable a wide range of applications, including text generation, classification, sentiment analysis, information retrieval, and content summarization [33,34,35].

In the field of marketing, consumer reviews provide critical insights for optimizing products and services. The abundance and accessibility of consumer review data offer a solid foundation for researchers to employ LLMs in text mining studies. For instance, Praveen et al. fine-tuned four major open-source LLMs (BERT, GPT-2, MPT-7B, and Falcon-7B) on consumer review data and enhanced their performance through prompt engineering, thereby demonstrating the remarkable capability of LLMs in consumer review topic modeling [36]. Bikku et al. applied BERT to a social media dataset for ternary sentiment analysis and found that BERT significantly outperformed traditional approaches in sentiment analysis tasks within the social media context [37]. Similarly, Li et al., using open-ended questionnaire data from consumers, showed that compared with conventional clustering methods, LLMs-based workflows enhanced both consumer segmentation and preference identification [38]. Collectively, these studies highlight the versatility of LLMs in extracting topics from unstructured text, improving clustering, and capturing nuanced sentiment.

More recent literature has further underscored the significance of LLMs in text mining. Viswanathan et al. proposed an iterative clustering technique in which LLMs are employed to refine and correct clustering outcomes, particularly by evaluating and reassigning low-confidence data points to more appropriate clusters, thus improving both accuracy and efficiency [39]. In another study, Liu et al. integrated LLMs with graph convolutional networks (GCNs) to develop a hybrid model addressing semantic drift and implicit sentiment issues in aspect-based sentiment analysis, thereby enhancing the granularity of sentiment detection [40]. Consequently, in the domain of urban renewal, there is an urgent need to advance research that leverages resident comments—often more complex than consumer reviews—for the analysis of resident satisfaction.

3. Data and Methodology

This study proposes a LLMs–based analytical framework for assessing residents’ satisfaction with urban renewal in old communities. The framework employs LLMs to automatically conduct semantic analysis of residents’ online messages and to extract keywords that influence satisfaction. These keywords are then categorized into thematic dimensions in accordance with residents’ satisfaction theory, as illustrated in Figure 1.

3.1. Data Sources and Preprocessing

Against the backdrop of urban regeneration, Beijing, as one of China’s megacities, has set the goal of becoming a “world-class, harmonious, and livable city.” Due to its unique political, economic, and social status, the city has consistently been at the forefront of China’s urban renewal practices [41]. Since the launch of a large-scale renovation of old residential communities in 2019, Beijing’s core districts—Dongcheng and Xicheng—have served as key sites, completing renovation projects in 183 communities by 2023. Simultaneously, the “Leadership Message Board” on People’s Daily Online represents a concrete practice of socialist democracy with Chinese characteristics. All Chinese citizens, either in their individual capacity or as legal entities, may undergo real-name authentication, select the relevant region and corresponding government officials, submit questions and appeals, and then file them formally. Operated by central media institutions, including People’s Daily and People’s Daily Online, this platform functions as a channel of public inquiry and governmental response. Its operational model ensures openness while simultaneously safeguarding order and effectiveness in communication between the public and the authorities.

Accordingly, this study employed Python software (python 3.12) to collect residents’ comments from the People’s Daily Online Leadership Message Board. Using “urban governance” as the primary keyword, combined with restrictive terms such as “old residential communities,” “community renovation,” and “community renewal,” and setting the time frame from 28 February 2024, to 31 December 2024, a total of 2502 comments from the Beijing region were retrieved. Traditional text mining typically requires extensive preprocessing of textual data. In contrast, the present study applies LLMs for text analysis, which greatly simplifies the preprocessing stage. Only the following steps were applied to the collected residents’ comments: (1) Removal of comments concerning city-level. Such messages pertain to municipal engineering projects and fall outside the scope of community-level renewal [42]. (2) Removal of irrelevant comments unrelated to residents’ willingness or attitudes toward community renewal. A significant portion of these were migrant workers’ complaints regarding wage arrears in renovation projects, which are beyond the scope of this research. (3) Removal of comments with fewer than ten characters. These messages generally lack sufficient information and are unsuitable for topic identification or sentiment analysis. After applying these filtering procedures, the final dataset comprised 2006 valid resident comments.As shown in Figure 2.

3.2. Topic Identification with LLMs

In this study, LLMs were integrated with Python to conduct interactive analysis of collected posts, thereby enabling large-scale, automated text processing. In prior research, topic modeling has frequently relied on methods such as LDA and Probabilistic Latent Semantic Analysis (PLSA). However, these approaches often neglect semantic relationships among words and perform inadequately in capturing co-occurrence patterns within documents, particularly when dealing with noisy, short-text formats [17,43,44]. The objective of this study is to leverage the unsupervised learning capacity of LLMs to identify the inherent semantics and sentiments embedded in residents’ comments.

Within the dataset of 2006 resident comments, each entry may encompass multiple factors influencing resident satisfaction, rendering the analysis more complex than aspect-based evaluations of consumer reviews in e-commerce contexts. To address this, the present study first employs the large-scale knowledge base inherent in LLMs to conduct vectorized analysis of each resident comment individually. Subsequently, logical chains reflecting the factors that affect resident satisfaction are extracted. This process ensures that the LLMs can filter out irrelevant information, structure the content, and refine the logical progression of residents’ statements. Finally, from the reconstructed logical chains, keywords are identified and a satisfaction score ranging from –1 to 1 is assigned. The results are presented in

Table 2. Keyword and satisfaction identification process.

Resident Message	Logical Chain	Keywords	Satisfaction
In the comprehensive renovation project of the old residential community at Haite Garden West District, Apple Orchard Subdistrict, Shijingshan District, the water supply and drainage renovation could not be carried out due to certain residents’ reasons. It is hoped that the leadership will provide assistance, pay attention to and respond to the demands of the majority of residents, and promote the refined advancement of the old community renovation project.	Opposition or lack of cooperation from certain residents (e.g., property disputes, individual demands) prevents the implementation of the water supply and drainage renovation → project stagnation as a whole → the reasonable demands of the majority of residents for improved living conditions remain unmet → dissatisfaction and disappointment arise → residents appeal to leadership for assistance in advancing the project with greater refinement, thereby preventing a minority from obstructing public interests.	resident non-cooperation, construction stagnation, unmet demands, dissatisfaction, appeal to leadership	−0.5
In Jiugong Town, a large number of old residential communities, mostly built in the 1990s, have accommodated the first generation of residents for nearly 30 years. Many households now belong to the elderly population. To improve the living experience of older adults, the installation of elevators in these old communities has become imperative.	Communities built in the 1990s, with residents living for nearly three decades, have experienced significant aging of the population → lack of elevator facilities makes it difficult for elderly residents to move between floors, diminishing their living experience → reduced convenience leads to lower resident satisfaction.	old residential communities, 1990s construction, aging residents, elderly living experience, elevator installation	−0.3

.

From the 2006 resident comments, a total of 8123 keywords were identified. Some of these appeared only once, primarily due to high semantic similarity among many keywords. For example, expressions such as “elevator aging,” “elevator maintenance,” “installing elevators,” and “retrofitting elevators” essentially convey the same meaning. To more accurately capture collective semantic representations and reflect residents’ core concerns, further analysis was conducted using LLMs.

First, the 8123 identified keywords were semantically vectorized and subjected to clustering analysis. Keywords with similar meanings were consolidated into thematic terms, and their frequencies were calculated. This process resulted in 119 thematic terms, which were then arranged in descending order by frequency. Given that most thematic terms had relatively low frequencies, the top 19 terms (accounting for approximately 61% of all occurrences) were selected for in-depth analysis. These high-frequency themes collectively represent the central issues raised by residents in their comments. Among them, the top three were: lack of transparency in policy information, inaction of property management/community committees, and drainage and safety concerns in Figure 3.

3.3. Explanatory Model for Residents’ Messages on Government Interactive Platforms

Current research on resident satisfaction with community renewal largely focuses on residents’ perceptions of different aspects of the renovation. Kang Lei et al. identified additional key determinants—including service facilities, job accessibility, and the built environment—thus enriching the understanding of satisfaction in terms of physical dimensions [45]. Chen et al., through an empirical comparison of pre- and post-renovation satisfaction, examined the mediating role of community management organizations (including neighborhood committees and property management) during the renewal process [46]. Xue et al., adopting a perspective of construction organization and resident–contractor interactions, analyzed how construction schedules, information disclosure, and conflict resolution affect residents’ participation attitudes and satisfaction [47]. As research progresses, factors such as neighborly relations, sense of community attachment, and resident expectations have also been found to be related to resident satisfaction [48,49]. In summary, the existing literature identifies three major dimensions influencing resident satisfaction. First, satisfaction with the perceived renovation outcomes of community renewal, typically reflected in housing conditions, the built environment, and infrastructure. Second, the perceived renovation interactions among different stakeholders during the renewal process—primarily among residents themselves and between residents and community organizations—play a critical role in shaping perceptions of satisfaction. Third, the perceived renovation procedures aspects of construction governance are equally vital, as they significantly affect residents’ perceived satisfaction with the renewal outcomes.

Accordingly, this study constructs a theoretical model of resident satisfaction in the context of old community renewal, encompassing three dimensions: residents’ perception of renewal outcomes, residents’ perception of renewal-related interactions, and residents’ perception of renewal procedures in Figure 4.

Based on the constructed theoretical model, this study analyzes the 19 thematic terms extracted from residents’ comments.

First, themes such as water supply and drainage pipelines, insufficient heating, aging housing, safety hazards, parking problems, elevator installation, poor environmental conditions, traffic congestion, and noise disturbance reflect residents’ perceptions of renovation outcomes, which are the primary determinants of satisfaction. Among them, the first four relate to basic living needs and thus correspond to fundamental renovations, while the latter four aim to enhance quality of life and therefore represent upgrading renovations.

Second, themes including perceptions of unfairness, unmet demands, communication deficiency, ineffective complaints, inaction of property management/community committees, and ineffective neighbor communication capture residents’ perceptions of interactions among stakeholders during the renewal process, which are critical to satisfaction. Specifically, perceptions of unfairness and unmet demands reveal shortcomings in the early-stage collection of residents’ opinions, where needs were not comprehensively or accurately addressed. communication deficiency, ineffective complaints highlight the absence of effective channels for residents to participate in decision-making and safeguard their rights, underscoring deficiencies in mechanisms of resident participation. Meanwhile, inaction of property management/community committees together with ineffective neighbor communication reflects insufficient coordination and conflict resolution mechanisms among multiple actors, pointing to communication-related issues within the community.

Finally, themes such as policy information non-transparency, severe information deficiency, construction quality issues, and living inconvenience represent residents’ perceptions of construction procedures, which are also crucial to satisfaction. Policy information non-transparency and severe information deficiency gaps stem from inadequate publicity and disclosure by relevant departments, making it difficult for residents to access or interpret key information, and thus reflecting insufficient policy transparency. Conversely, construction quality issues and living inconvenience reveal weaknesses in quality management, schedule control, and civilized construction practices, which cause significant disturbances to residents’ daily lives, highlighting shortcomings in construction governance. The results are summarized in

Table 3. Variable definitions.

Theme	Variable Name	Variable Explanation
Enhancement- Oriented Renovation	Parking Problems	Severe shortage or disorderly management of parking spaces within the community, resulting in difficulties in parking and widespread illegal parking
	Elevator Installation	Installation of elevators in old multi-story residential buildings to address mobility challenges for upper-floor residents, particularly the elderly.
	Poor Environmental Conditions	Accumulation of garbage, poor sanitation, unauthorized constructions, lack of greenery, or inadequate maintenance in community public areas.
	Traffic Congestion	Narrow internal or entrance/exit roads, mixed pedestrian–vehicle traffic, and on-street parking causing slow movement and congestion.
	Noise Disturbance	Construction noise or excessive community noise (e.g., square dancing, home renovations) exceeding reasonable limits and affecting residents’ rest.
Basic Renovation	Water Supply and Drainage Pipelines	Aging, clogged, or damaged water supply and drainage systems within the community.
	Insufficient Heating	Winter heating that fails to meet standards or is unevenly distributed, leading to excessively low indoor temperatures.
	Aging Housing	Deterioration of residential buildings due to age, including structural wear, wall detachment, and outdated facilities.
	Safety Hazards	Risks within the community such as dilapidated walls, blocked fire escapes, aging electrical wiring, or lack of security that may endanger life and property.
Resident Intention Solicitation	Perception of Unfairness	Residents perceive inequities in the allocation of renovation resources, compensation schemes, or policy implementation, resulting in perceived losses of personal interests.
	Unmet Demands	Specific opinions, suggestions, or requests raised by residents during the renovation process are not effectively addressed or resolved.
Resident Decision-Making Power	Communication Deficiency	Lack of proactive, sufficient, and two-way information exchange between residents and stakeholders (government, implementing units, community committees/property management).
	Ineffective Complaints	Complaints raised by residents regarding renovation issues fail to receive effective handling or substantive solutions.
Policy Transparency in Renovation	Policy Information Non-Transparency	Key information concerning renovation policies, construction progress, decision-making processes, and fund usage is not clearly disclosed to residents.
	Severe Information Deficiency	Residents face difficulties in accessing detailed information regarding renovation planning, progress, policies, or compensation standards.
Construction Governance	Construction Quality Issues	Substandard construction caused by corner-cutting, rough workmanship, or the use of non-compliant materials during renovation projects.
	Living Inconvenience	Disruptions to daily life (e.g., shopping, commuting) due to facility suspensions or altered routes during or after renovation.
Community Communication	Inaction of Property Management/community Committees	Failure of property management companies or community committees to effectively coordinate, maintain order, facilitate communication, or perform routine services during renovation.
	Ineffective Neighbor Communication	Difficulty in reaching consensus or engaging in effective negotiation among residents or between residents and other stakeholders regarding renovation matters.

.

In this research, the contextual “memory capacity” of LLMs is utilized to assign keywords contained within each sentence. This assignment produces a binary 0/1 variable, indicating whether a specific keyword is present in the sentence. As discussed earlier, the main keywords are categorized into seven thematic dimensions: basic renovation, improvement-oriented renovation, collection of residents’ preferences, residents’ decision-making power, policy transparency, construction governance, and community communication.

Specifically, if a resident’s message contains references to themes such as parking issues (T10), elevator installation (T11), poor environmental conditions (T8), noise disturbance (T14), or traffic congestion (T18), the LLMs assigns a binary indicator: 1 if any of these themes are mentioned, and 0 otherwise. This process results in the construction of a topic matrix. Table 2 presents the thematic classification and variable explanations derived from the resident satisfaction theoretical model.

Given the strong explanatory power, predictive capability, and relative simplicity of multiple regression models in multifactor analysis, this study employs the satisfaction [50] score of each resident comment (i) as the dependent variable. The explanatory variables include Enhancement-Oriented Renovation, Basic Renovation, Resident Intention Solicitation, resident decision-making power, policy transparency in Renovation, construction governance, and community communication. Accordingly, a multiple regression model is constructed, with the final equation expressed as (1).

To prevent potential collinearity among variables that might bias the model results, variance inflation factor (VIF) tests were conducted for all explanatory variables. As shown in

Table 4. Variance inflation factor.

Theme	Enhancement-Oriented Renovation	Basic Renovation	Resident Intention Solicitation	Resident Decision-Making Power	Policy Transparency in Renovation	Construction Governance	Community Communication
VIF	1.05	1.08	1.22	1.48	1.09	1.06	1.42

, all VIF values are below 2, indicating the absence of multicollinearity.

Rating = β₀ + β₁ Improvement oriented Renovation_i + β₂Basic Renovation_i + β₃Residents Willingness Collection_i + β₄Resident Decision making Power_i + β₅Policy Transparency_i + β₆Construction Governance_i + β₇Community Communication_i + a_i

(1)

4. Results

4.1. Comparative Analysis of LLMs and LDA Results

To compare the benefits of LLMs in semantic understanding for topic clustering, this study also employs the (LDA) topic modeling method to analyze tweets for comparative research. LDA is a widely applied statistical unsupervised machine learning technique used to identify latent thematic information from large-scale corpora. During the preprocessing stage, residents’ comments were segmented using the Jieba tool (Jieba 0.72). A combination of a general Chinese stop-word list and domain-specific noise words (e.g., “leader,” “hello,” “thank you,” and punctuation) was applied to remove meaningless terms, while retaining substantive words such as “elevator,” “leakage,” and “illegal construction.” To determine the optimal number of topics, topic coherence and perplexity were evaluated across different settings, with the number of topics selected at the point where coherence was maximized and the decline in perplexity began to level off [51]. The number of iterations was set to 1500 to ensure sufficient model convergence [52]. Finally, LDA modeling was performed on the corpus of resident comments, yielding both the topic distribution for each comment and the global topic–word probability distribution.

Since the LDA method merely clusters Chinese words, the latent topics require manual labeling. Therefore, in this study, the clusters were manually annotated with the most appropriate topic labels, informed by the analytical results of the LLMs, in order to best explain the combinations of keywords. The results in

Table 5. LDA topic analysis.

Theme	Keywords
Transportation Infrastructure	Renovation Residents Roadways Severe Environment Property Management Impact Habitation Property Owners Vehicles Construction Street Enhancement Planning Inability Retrofitting
Administrative Organization	Property Property Owners Residents Temperature Demands Property Management Maintenance Heating System Situation Street Property Management Company Real Estate Developer Property Management Fees Complaints
Waterproofing Infrastructure	Renovation Residents Water Leakage Rainwater Leakage Severe Repair Old Housing Renovation Redevelopment Housing Street Conduits Households System Aged Residential Living Waterproofing Situation Plan Heating Supply Thermal Insulation Wall Structure As Soon as Possible Environment Inability Installation Additional Installation Water Pipes Insulation
Economic Compensation	Residents Renovation Demolition and Relocation Construction Environment Planning Improvement Property Owners Residential Living Roadways Additional Installation Plan Urban Development Chaoyang (District/Area) Construction Work Rectification
community Relations	Residents Additional Installation Property Owners Street Property Management Kindergarten Renovation Residential Living Construction Impact Parking Spaces Housing Old Housing Renovation Redevelopment Inability Consent Situation Installation

show that within the same topic, the relationship between keywords and their corresponding themes is often unclear, making it difficult to discern the specific semantic expressions of residents concerning the renovation of old communities—for instance, the appearance of words such as “temperature” and “heating” under the theme “management organization.” Moreover, there is a high degree of keyword overlap across different topics, further complicating thematic interpretation. In contrast, as demonstrated in Table 2, topic word extraction and sentiment analysis by the LLMs yield more reasonable and accurate results, with superior interpretability that aligns more closely with human judgment. Hence, the analytical results produced by LLMs exhibit greater interpretability and accuracy.

4.2. Analysis Results of Resident Satisfaction

During the process of community renovation, residents’ satisfaction is influenced by multiple factors. A multiple linear regression model enables simultaneous consideration of the effects of multiple independent variables (predictors) on a single dependent variable (response). Accordingly, this study constructs a multiple linear regression model, with the regression results presented in

Table 6. Multiple regression results.

Var	u
Enhancement-Oriented Renovation	−0.046 ***
	(−7.506)
Basic Renovation	−0.075 ***
	(−11.772)
Resident Intention Solicitation	−0.166 ***
	(−21.300)
Resident Decision-Making Power	−0.021 ***
	(−2.654)
Policy Transparency in Renovation	−0.017 **
	(−2.020)
Construction Governance	−0.065 ***
	(−7.496)
Community Communication	−0.158 ***
	(−22.052)
adj. R2	0.496
N	2006

t statistics in parentheses; ** p < 0.05, *** p < 0.01.

.

(1)

Perceived renovation outcomes (basic and enhancement-oriented renovations) on Resident Satisfaction

The model results indicate that deficiencies in fundamental renovations (−0.075, p < 0.01) and upgrading renovations (−0.046, p < 0.01) are both significantly negatively correlated with resident satisfaction. Moreover, when residents perceive stronger deficiencies in fundamental renovations—such as inadequate heating, deteriorated housing, or outdated water and sewage pipelines—the negative impact on satisfaction is greater than that associated with upgrading deficiencies such as parking shortages or the absence of elevators.

This finding suggests that in the context of Beijing’s old community renewal, shortcomings in fundamental renovations—which are closely tied to residents’ basic living needs and safety—substantially undermine residents’ recognition of renovation outcomes. Although dissatisfaction stemming from the absence of upgrading renovations is weaker compared to fundamental issues, residents nonetheless view them as important signals of whether their quality of life can be effectively improved.

(2)

Perceived renovation interactions (resident participation) on Resident Satisfaction

Within the dimension of residents’ perceived interactions, both early-stage opinion solicitation (−0.166, p < 0.01) and mid-to-late stage participation in decision-making (−0.021, p < 0.01) show significant negative correlations with satisfaction. This indicates that different forms of resident participation throughout the renewal process are significantly associated with lower satisfaction. Among these, opinion solicitation exerts the greatest impact, suggesting that in Beijing’s old community renewal projects, existing channels for collecting residents’ opinions are inadequate and feedback mechanisms are ineffective, leading to strong perceptions of unfairness. This may be linked to the complexity of property rights and resident composition in Beijing’s old communities. In contrast, although mid-stage participation in decision-making is important, its relative impact is more limited.

Community communication also emerges as a key determinant of satisfaction. The results show that inefficient communication among neighborhood committees, property management, and residents (−0.158, p < 0.01) is significantly negatively correlated with satisfaction, second only to opinion solicitation. Ineffective communication amplifies residents’ dissatisfaction: in practice, Beijing’s old community renewal projects often rely on grassroots organizations such as neighborhood committees and property managers, whose inaction, coupled with strained communication among residents, exacerbates conflicts. This highlights that residents’ disappointment with “people” (management absence, poor communication) outweighs their dissatisfaction with “things” (construction quality) or “procedures” (lack of transparency). It underscores that responsive daily services are more critical to resident satisfaction than hardware improvements alone.

(3)

Perceived renovation procedures (renovation procedures) on Resident Satisfaction

Within renovation procedures, both low policy transparency (−0.017, p < 0.05) and poor construction management (−0.065, p < 0.01) show a degree of negative correlation, with construction quality having a more substantial impact on satisfaction. Construction quality represents the baseline threshold for residents’ satisfaction; deficiencies (e.g., substandard materials, poor workmanship) directly provoke strong dissatisfaction. In contrast, lack of transparency in renovation policies undermines trust (e.g., failure to disclose project plans in a timely manner), yet its impact is only one-quarter that of quality-related issues. Therefore, governance should prioritize stringent quality control standards (e.g., third-party inspections and acceptance), while transparent communication (e.g., public disclosure of construction progress) can help mitigate secondary conflicts.

5. Conclusions

5.1. Summary

With economic and social development, residents’ housing needs are becoming more sophisticated, and understanding resident satisfaction has become more complex. This study combines text analysis with resident satisfaction theory to investigate the factors influencing resident satisfaction in old community renovations. The main research findings are as follows:

(1)

From the volume of resident messages, we found that the issues residents are most concerned about include: lack of transparency in policy information, inaction of property management/community committees, drainage and safety concerns, water supply and drainage pipelines, perception of unfairness, insufficient heating, ineffective complaints, poor environmental conditions, aging housing, parking problems, elevator installation, ineffective neighbor communication, noise disturbance, construction quality issues, living inconvenience, traffic congestion, communication deficiency, unmet demands, and severe information deficiency. Furthermore, compared to LDA, LLMs analysis results show higher interpretability and accuracy.

(2)

We then identified seven key influencing factors: Enhancement-Oriented Renovation, Basic Renovation, Resident Intention Solicitation, Resident Decision-Making Power, Policy Transparency in Renovation, Construction Governance, and Community Communication. While these are consistent with previous studies, we integrated these seven factors into a unified framework based on the theoretical dimensions of Perceived renovation outcomes, Perceived renovation interactions, and Perceived renovation procedures, thereby unifying the factors influencing resident satisfaction.

(3)

Finally, we conducted a multiple regression analysis using the sentiment values of residents’ comments as the dependent variable. The results highlight the critical role of resident intention solicitation in urban regeneration, suggesting that governments and planners should attach greater importance to investigating and analyzing residents’ intentions at the early stages of renewal projects. In addition, difficulties in community communication exert a strong negative effect on resident satisfaction, indicating that grassroots organizations such as neighborhood committees must improve mechanisms for mediating communication between residents and other stakeholders. With regard to the differential impacts of enhancement-oriented renovation and basic renovation, the findings suggest that most communities should follow a rational prioritization: placing basic renovations that address essential living needs at the forefront, while treating enhancement-oriented measures as supplementary. Lastly, the results reveal that residents express relatively low dissatisfaction with policy transparency during the construction phase, but exhibit pronounced dissatisfaction with construction governance. This indicates that while progress reporting has been relatively effective, substantial gaps remain in quality control, falling short of residents’ expectations.

5.2. Contribution

(1)

Methodologically, this study applies LLMs to advance the analysis of residents’ feedback texts in the field of old community renovation. An innovative approach is adopted, leveraging the semantic understanding and generative capabilities of LLMs to extract keywords and emotions related to residents’ satisfaction from their messages. These keywords are then clustered and interpreted. Compared with traditional LDA-based topic modeling, this approach introduces a novel and more objective method for determining topic labels and analyzing hierarchical topic structures. Ultimately, by employing LLMs, the study enhances the objectivity and depth of text mining, thereby addressing the subjectivity inherent in conventional topic labeling.

(2)

Theoretically, this study contributes to the development of resident satisfaction theory. Existing research on satisfaction in the context of old community renovation has often focused on single-dimensional factors such as housing conditions, built environment, community organizations, or neighborhood relations. In contrast, this study integrates theories of resident satisfaction with textual analysis of residents’ feedback, proposing a more systematic explanatory model to consolidate the key factors affecting satisfaction in community renovation. Specifically, the study organizes these factors into three dimensions—residents’ perceptions of renovation outcomes, perceptions of renovation procedures, and perceptions of renovation interactions—thus providing a unified framework that expands and refines the existing theoretical landscape.

5.3. Managerial Implications

The research findings provide actionable insights for governments and third-party enterprises involved in community renovation, enabling them to better serve residents and advance urban renewal.

(1)

Within the of old communities Regeneration in Beijing, foundational improvements must remain the primary focus, serving as the core direction for both policy and fiscal support. Meanwhile, supplementary enhancement projects are considered a crucial complement. In instances where a community’s basic infrastructure is already robust, a strategic shift should prioritize high-impact enhancement projects to optimize renovation outcomes. Specifically, this entails the prioritized installation of elevators in neighborhoods with a large population of senior citizens and children, along with the deliberate allocation of public space for parking to address resident demand.

(2)

Local regulations, such as the Beijing Urban Renewal Ordinance, explicitly stipulate that redevelopment must “fully solicit residents’ opinions” and safeguard their “right to information, participation, and decision-making.” However, research indicates that in practice, the implementation of these rights remains inadequate, particularly in the early stages of renewal, owing to the heterogeneity of old community residents in terms of locality, age, gender, and other characteristics. This often results in insufficient opinion solicitation and limited realization of participation rights. Therefore, it is essential to expand participation channels and refine methods of soliciting residents’ opinions. Collaborating with reliable third-party organizations in community renewal, and leveraging artificial intelligence technologies to develop application functions that generate personalized content based on residents’ preferences and behaviors, constitutes a critical step toward achieving this goal.

(3)

In state governance, neighborhood committees are positioned as organizers, property management companies as both “investors and operators,” and homeowners’ associations or residents’ assemblies as final decision-makers. Incentive and sanction mechanisms aim to shift grassroots governance from short-term fixes to long-term stability within a unified framework of rights, responsibilities, and benefits. Yet without institutional refinement and oversight, the roles of committees and property managers may cause imbalance and erode trust. Thus, joint meetings, clear responsibility lists, third-party evaluations, and rapid-response mechanisms are needed to improve transparency and responsiveness to residents.