1 Introduction
In the context of globalization, cultural and creative industries have emerged as an important economic sector, contributing to the transformation of urban economic structures while reshaping cultural landscapes and social life
[1]. For urban regeneration practices worldwide, cultural and creative spaces have become an important part for the transformation of post-industrial cities. Cities such as London, New York, Seoul, and Berlin have transformed declining industrial areas into multifunctional spaces for cultural consumption and innovation through the adaptive reuse of industrial buildings, the clustering of creative industries, and the development of cultural and consumption-oriented environments
[2–
6]. Creative industry parks (CIPs hereafter), often developed through the revitalization of obsolete factories, warehouses, and other underused spaces, provide physical settings for the clustering and development of creative industries while also serving as important instruments of urban regeneration, spatial reuse, and cultural innovation.
Research on CIPs extends across multiple disciplines, including Architecture
[7], Geography
[8], Economics
[9], and Sociology
[10]. Previous studies have focused primarily on functional roles
[11–
12], spatial morphology
[13–
14], evolutionary processes
[15], interactions with local economies and cultures
[16], and strategies for regeneration, sustainable development, and spatial vitality enhancement
[17–
19]. As social media becomes increasingly integrated into urban experience, however, the attractiveness of CIPs depends not only on the quality of their physical environments but also on their online visibility, capacity for content circulation, and public evaluations on digital platforms
[5,
20]. Platforms such as Instagram, Google Reviews, Tripadvisor, and Dianping have become important channels through which users experience, evaluate, and share information about urban cultural spaces
[21–
22]. Users’ experiences of visiting, consumption and recreation are then translated into digital content, including comments, images, and ratings, which circulate and accumulate on platforms and, in turn, shape visitor expectations and inform CIP operations and management. Accordingly, researchers have increasingly used social media data to examine CIPs. For example, Yixin Zhang et al. used semantic analysis of social media data to identify factors associated with the popularity of CIPs
[20], while Luming Cui et al. examined how social platforms contribute to the restructuring of urban creative spaces using user-generated content (UGC)
[23]. More recently, advances in semantic modeling have improved the ability to identify sentiment patterns and prominent topics in large-scale textual datasets, creating new opportunities to examine the structure of public perceptions and their spatial implications
[24].
However, research on CIPs both in China and internationally continues to rely largely on traditional methods, including questionnaires, field observations, and case studies, with greater emphasis on spatial form and layout than on visitors’ actual experiences and perceptions. Although some studies have applied large-scale data analysis to examine patterns of spatial use in CIPs
[25–
26], systematic analyses of public sentiment and key concerns remain limited. To address these gaps, this study develops an analytical framework integrating online visibility assessment, sentiment analysis, and topic identification, and uses the resulting evidence to formulate spatial optimization strategies for CIPs. Focusing on selected CIPs in Shanghai, this study analyzed user-generated comments from Dianping using Senta for sentiment classification and BERTopic for unsupervised topic modeling based on Transformer embeddings. Comment count and the time elapsed since reopening were used to characterize each CIP’s online visibility, while sentiment and topic structures were analyzed to examine differences in public perceptions and the relationship between online representations and reported on-site experiences. The study thereby seeks to identify visitor needs and experience-related problems more precisely, inform efforts to reduce the homogenization of CIPs, improve spatial quality and operational performance, and provide evidence-based guidance for CIP regeneration and targeted management.
2 Spatial Restructuring of CIPs From a Social Media Perspective
This research aimed to examine how social media shapes the online place images of CIPs, and how online public feedback can inform improvements to physical space.
2.1 Shaping the Digital Presence of CIPs Through Social Media
Existing studies generally regard the online place image of a CIP formed on social media as a reflection jointly constructed through user experience records, platform content aggregation, and social dissemination. Through the continuous accumulation of such data, the public forms an overall impression of spatial quality, service level, and cultural atmosphere
[23,
27]. During platform dissemination, algorithmic recommendation, user interaction, and social sharing further amplify online visibility, allowing CIPs to transcend the limits of geographic proximity and influence the cognitive frameworks and potential visitors’ willingness
[28–
31]. At the research level, comment count, temporal distribution of comments, and fluctuations in attention can reflect the level of interest and dissemination intensity of CIPs on digital platforms; high-frequency elements in comment texts and images can reveal the spatial features that are repeatedly perceived and expressed by the public. However, existing research often conducts descriptive analyses and lacks in-depth characterization of the sentiment structures and topic compositions underlying online place images and their relationship with spatial governance.
2.2 Spatial Response Pathways of CIPs to Social Media
As online place images accumulate and spread on platforms, the focus of CIP governance increasingly shifts toward how to respond to online feedback and establish adjustment mechanisms. Studies indicate that the spatial response of CIPs takes a cycle of online expression, dissemination and aggregation, and offline adjustment: UGC transforms experiences into shareable evaluations, which spread through platform mechanisms and affect visitor expectations and consumption choices, thereby exerting constraints and incentives on CIP spatial design and facilities and operational governance
[32–
33]. Accordingly, existing studies have summarized response pathways mainly in terms of the configuration of recognizable spatial features, optimization of experience chains, and incorporation of platform evaluations into governance feedback, with particular attention to basic aspects such as transport accessibility, wayfinding organization, sanitation facilities, and service management in order to reduce the accumulation and diffusion of negative evaluations
[34]. Yet specific studies often lack comparable and traceable empirical translations of these pathways, especially statistical descriptions of digital presence and its expansion, making it difficult to reveal the complete process of attention generation and differentiated evolution of parks on online social media platforms. In addition, structured identification of public feedback remains insufficient. Existing research often discusses response pathways through individual cases or small samples and rarely uses large-scale UGC to reveal the sentiment orientations and topic structures of public evaluations, hardly linking spatial elements with sentiment outcomes or providing evidence for differentiated governance.
Accordingly, this study develops a measurable framework based on social-media UGC from three dimensions: digital presence assessment, sentiment analysis, and topic identification. First, statistical indicators such as cumulative comment count and the years of operation were used to characterize the intensity of digital presence and the expansion of public attention. Next, sentiment analysis quantified the sentiment orientation and temporal fluctuations of public evaluations, while topic modeling identified the spatial elements and issue types concerned by the public. On this basis, differences in public perceptions among CIPs were interpreted, and corresponding strategies for spatial optimization were proposed.
3 Research Methods
3.1 Study Area and Data Sources
Shanghai is one of China’s leading economic and cultural centers and a global metropolis. Its CIPs have expanded rapidly and become prominent sites of innovative urban regeneration. Since the government designated its first 15 CIPs in 2009, Shanghai has developed an extensive network comprising 149 municipal-level CIPs
[35].
This study employed users’ comments from Dianping. Dianping is a major local-services platform in China, where users commonly post comments after visiting and consuming at physical venues. These comments provide large-scale, detailed, and temporally continuous records of user experience. The dataset includes comment text, user ID, and publication timestamps, providing the information required for quantitative sentiment and topic analysis.
To ensure sufficient textual information and reduce random variation associated with small samples, 37 of the 160 CIPs included in the 2023–2025 List of Shanghai Creative Industry Parks were selected as the study samples by comment count (more than 200) (Fig. 1). User comments posted between January 2016 and July 2024 were collected using Python-based web scraping, yielding 69,643 raw records for subsequent analysis.
To improve the reliability of text analysis, the raw comments were systematically preprocessed. Specifically, jieba was used for Chinese word segmentation; the HIT stop-word list was applied to remove common stop words, and non-semantic elements such as whitespace, punctuation, and emojis were removed. Duplicate comments, advertisements, and entries unrelated to evaluation were also excluded. After preprocessing, 68,826 valid comments remained and were used in subsequent analyses (Table 1).
3.2 Research Framework
This study developed a quantitative framework comprising digital presence assessment, sentiment analysis, and topic identification (Fig. 2). Digital presence assessment captures a CIP’s online visibility and the accumulation and circulation of comments. Sentiment analysis and topic identification measure the orientation and content of public evaluations. Together, they describe online public perception
[36]. Examining the differences in public perception among CIPs can inform CIPs’ online place-image management, spatial improvement, and operations.
(1) Statistical characterization of digital presence and its expansion
Cumulative comment count measures the strength of a CIP’s digital presence. The number of years of operation captures the temporal development of online visibility. The analysis then compares the distribution and accumulation of comments in central and suburban Shanghai. Finally, CIPs are classified by function and character to support comparisons of sentiment and topic structures among CIP types.
(2) Sentiment recognition and temporal differentiation analysis
This study used the Senta model based on a CNN architecture to identify sentiment in comment texts
[37]. Compared with sequential models such as LSTM, CNN is more suitable for extracting local sentiment features through multi-scale convolution and provides higher parallel-computing efficiency, thereby supporting stable processing of large-scale short texts. For each comment, the model outputs a positive-sentiment probability ranging from 0 to 1; higher scores indicate a stronger positive tendency. Following the sentiment-score classification method used in previous research
[38], comments with scores of 0.6 or higher were classified as positive, those with scores of 0.4 or lower as negative, and those between 0.4 and 0.6 as neutral. Key training and hyperparameter settings included convolution-window sizes of 2, 3, and 4; max pooling in the pooling layer; a dropout rate of 0.5 in the fully connected layer; batch-gradient-descent training with a batch size of 32; an initial learning rate of 5e-5 with decay across epochs; and a 7:2:1 split of the data into training, validation, and test sets. The number of training epochs was controlled according to validation performance to mitigate overfitting. Testing showed that the Senta model achieved 89.2% classification accuracy and an F
1 score of 0.87 on the study dataset, demonstrating its applicability in this research. Sentiment scores were further aggregated by time to characterize the overall sentiment orientation and temporal fluctuations of public evaluations.
(3) Topic identification and topic-structure analysis
This study used BERTopic for unsupervised topic mining of comment texts to characterize topic structures and identify the spatial elements and issue types that attract public attention
[39–
41]. Compared with bag-of-words models such as LDA, BERTopic is more suitable for short-text corpora. On the one hand, UGC texts are sparse and contain abundant synonymous expressions, making topic inference based on word-frequency co-occurrence sensitive to lexical variation. On the other hand, the actual number of topics is difficult to predefine and the corpus contains noisy texts, requiring adaptive clustering and noise-recognition capabilities. BERTopic represents sentence-level semantics through Transformer-based semantic embeddings and combines density-based clustering to form topic clusters, allowing stable topics to be identified and noise to be filtered without pre-specifying the number of topics. The workflow includes semantic embedding, dimensionality reduction, clustering, and keyword extraction. First, semantic embedding vectors were generated with a pretrained Transformer language model. UMAP was then used for dimensionality reduction to preserve the main semantic structure while reducing computational complexity, followed by adaptive clustering using HDBSCAN. Finally, topic keywords were extracted using class-based weighting and the topics were named accordingly. Key parameters were an embedding dimension of 384, a post-UMAP dimension of 15, a minimum cluster size of 10, and a c-TF-IDF weighting coefficient of α = 0.75. The model outputs were manually reviewed to ensure semantic coherence and interpretability, and a hierarchical diagram was used to visualize topic distributions and interrelationships. Following the overall topic analysis, positive and negative comments were modeled separately with BERTopic as two independent corpora. The resulting topic structures and representative keywords were compared to identify the dominant themes associated with each sentiment category and to examine their corresponding experiential and spatial implications.
4 Research Results
4.1 Statistical Characterization of Digital Presence and Its Expansion
Regarding digital presence, the comment counts of the selected CIPs showed a pronounced hierarchical distribution (Fig. 3). Qibao Old Street Folk Base ranked first with 24,277 comments, followed by Columbia Circle, 1933 Old Millfun, Shanghai Museum of Glass, and Shanghai Fashion Center. At the same time, digital presence does not increase linearly over time but is associated with stages of spatial regeneration (Fig. 4). The regeneration of the selected CIPs was concentrated mainly between 2010 and 2016. Notably, although Columbia Circle opened at the end of 2016, its comment count already ranked third, suggesting that highly visible CIPs can accumulate comments rapidly over a short period through a reinforcing sequence of visits, comments, and sharing.
Spatially, the selected CIPs are concentrated in central Shanghai, with comparatively few located in suburban areas. Beyond differences in location and accessibility, this concentration may also be associated with the greater intensity of everyday activities, consumption opportunities, and social media use in central areas, which can facilitate more frequent and sustained comment accumulation. To examine differences in public perceptions across CIPs with different functions, the 37 CIPs were classified into five types: industry-focused, arts-oriented, leisure-and-entertainment, local heritage and culture, and mixed-use. Industry-focused CIPs primarily accommodate enterprises and industry-related services (e.g., Zhongcheng Wisdom Valley); arts-oriented CIPs emphasize artistic production and exhibition, often through the adaptive reuse of former factories and warehouses (e.g., M50 Creative Park, Columbia Circle); leisure-and-entertainment CIPs prioritize consumption and recreational experiences (e.g., Bridge 8, 800 Show, and The Modern); local heritage and culture CIPs emphasize historic character and local traditions (e.g., Jinshanzui Fishing Village, Qibao Old Street Folk Base, Qingxi Old Street); and mixed-use CIPs integrate industry services, leisure and entertainment, heritage features, and other functions (e.g., 1933 Old Millfun, Deqiu Li Xiang Valley).
4.2 Sentiment Analysis of CIPs
The overall mean sentiment score of the sampled CIPs was 0.83. As shown in Fig. 5, the mean score of each CIP was above 0.7, indicating that public evaluations of Shanghai’s CIPs were generally positive. There were 58,625 positive comments, accounting for 85% of the total; negative comments accounted for 10%, and neutral comments for 5%. Among the CIPs, M50 Creative Park and Happy Lane both had a mean sentiment score above 0.9, indicating relatively high levels. The higher-scoring CIPs were concentrated on the arts-oriented and leisure-and-entertainment ones, which often feature distinctive artistic characteristics, a leisure atmosphere, and strong visual identities, making them more likely to receive positive public evaluations.
Annual mean sentiment scores of all the studied CIPs increased from 2016 to 2019, declined substantially from 2020 to 2022, and gradually recovered after 2023 (Fig. 6). Before the COVID-19 pandemic, public evaluations improved steadily, suggesting that policy support and continued regeneration were associated with favorable assessments of spatial quality and operations. The decline during the pandemic might be mainly due to the reduced visitor activity, operating restrictions, and poorer public-service experiences. Scores recovered as in-person activities and operations resumed, possibly because of re-dynamics, improved facilities and services, and enhanced management.
4.3 Core Topics Analysis of CIPs
Based on the BERTopic model, 28 topics were identified from the comment texts and ranked by frequency (Fig. 7). The most frequent topics centered on everyday experiences, including child-friendly activities, transportation, historic buildings, dining, and shopping. The topics formed distinct clusters in semantic space, reflecting differences in the content of public discussions. Children recreation (Topic 0) was the largest cluster, suggesting that family leisure and parent–child experiences constitute a major focus of public discussion. Transportation (Topic 1) also received substantial attention, underscoring the importance of accessibility in public evaluations. Historical culture (Topic 2) and architectural heritage (Topic 3) both related closely to historic buildings, suggesting that architectural character and adaptive reuse are important components of how the public perceive and recognize CIPs. In addition, weekend dates (Topic 15) and weekday conditions (Topic 18) captured differences between weekend leisure and weekday visitation, revealing a distinct temporal pattern in park use. Topics related to social-media-popular restaurants, music performances, wedding celebrations, and animation exhibitions highlighted the diversity of park attractions and the importance of event-based experiences. At the same time, some topics pointed directly to deficiencies in infrastructure and visitor experience, suggesting that public services and operational management also play an important role in shaping overall perceptions.
Based on semantic similarity, the 28 topics were further organized into three thematic clusters: cultural and sports activities, leisure and daily life, and history and culture. This grouping revealed the semantic relationships and hierarchical structure of public comment content (Fig. 8). The cultural and sports activities cluster encompasses exhibitions, performances, celebrations, and sporting events; the leisure and everyday life cluster includes dining, photography and social media check-ins, nighttime scenery, and visits at different times; and the history and culture cluster centers on historic buildings, architectural heritage, and distinctive artistic features, while also incorporating practical aspects of the visitor experience such as transportation, recreation, and shopping.
4.4 Core Topics Analysis of Positive and Negative Comments
Positive and negative comments were modeled separately, revealing distinct differences in their thematic distributions (Fig. 9). Sentiment is primarily associated with three dimensions of the visitor experience: transport accessibility, attractive spatial features (e.g., historic buildings, art exhibitions), and the quality of public facilities (e.g., sanitation, service, park and crowd management). Positive comments center on accessibility, children recreation, art exhibitions, photography and social media check-ins, and historic landmarks. Favorable evaluations were more commonly associated with the combination of convenient access, distinctive spatial features, and engaging activity programming. By contrast, negative comments were dominated by expressions of disappointment and often compare current conditions with earlier experiences, indicating a perceived decline in visitor experience. These comments mainly concerned poor service attitudes, adverse weather conditions, and inadequate sanitation facilities.
The comparison suggests that spatial features with high potential for social media sharing, such as historic buildings, art exhibitions, and visually distinctive locations, are more likely to generate UGC and circulate online, contributing to positive perceptions. By contrast, basic service and management conditions, including sanitation facilities, service quality, and site management, may attract less attention in routine social media sharing. However, when deficiencies occur, they are more likely to appear collectively in negative comments and adversely affect overall perceptions of the CIP. Therefore, differences in public perception reflect not only variations in location and type but also the interaction between internal spatial provision, management quality, and the ways these characteristics are represented and circulated through social media.
5 Discussion
5.1 Mechanisms Underlying Spatial Differences of Public Perception
The research findings revealed that differences in public evaluations of CIPs cannot be attributed to CIP type or location alone; rather, they emerge through interactions between social media and visitors’ experiences of the physical space. Visitors selectively document spatial features and experiences through UGC, while platform recommendations and information flows shape the visibility and circulation of the content. The resulting digital presence of the CIP may shape potential visitors’ expectations and visitation decisions and is subsequently reinforced, modified, or challenged through later visits and newly generated content. These differences of public perception can be understood through three interrelated processes: the production and circulation of digital place representations, the role of basic service thresholds, and the uneven distribution of online attention and repeated representation of place.
First, visually distinctive and narratively engaging spatial features facilitate the production and circulation of digital place representations. On social media, these features can be readily communicated through photographs, narratives, and other UGC forms. This may help explain why arts-oriented CIPs and some leisure-and-entertainment CIPs received relatively favorable evaluations. Their distinctive settings can be documented, shared, and recirculated, allowing individual experiences to contribute to broader public expectations. Spatial design and activity programming therefore contribute to both on-site experience and the CIP’s digital presence, potentially reinforcing a cycle of visits, comments, and subsequent sharing.
Second, basic services establish a minimum threshold for a satisfactory visitor experience. Transport accessibility, sanitation facilities, service quality, and crowd management may receive limited attention when they meet expectations, whereas deficiencies in these areas are more likely to be highlighted in negative comments. When operational or maintenance problems persist, visitors may perceive them as evidence of broader deterioration in park quality rather than as isolated shortcomings. Positive online attention is therefore more closely associated with shareable spatial features, whereas negative evaluations are often linked to deficiencies in basic services and management.
Third, digital platforms reshape the distribution of attention and contribute to the repeated representation of place. A CIP’s digital presence is not determined solely by geographic proximity. The circulation, aggregation, and recommendation of UGC can extend the visibility of some CIPs beyond their immediate surroundings and may translate online attention into physical visits. Highly visible CIPs may attract more visitors and generate additional UGC, further strengthening their digital presence. CIPs with lower visibility or a less distinctive identity may be less likely to benefit from this cumulative process. Central-city CIPs may gain cumulative advantages from higher levels of activity, consumption, and social media use, whereas suburban CIPs may receive less online attention and less favorable evaluations when they lack a distinctive identity or adequate basic services. CIPs therefore derive their attractiveness from the combined quality of their physical environments, digital presence, and capacity to meet visitor needs.
5.2 Spatial Optimization Strategies for CIPs From a Public-Perception Perspective
Based on the sentiment and topic analysis of Dianping UGC, this study proposes the following spatial strategies and practical measures for CIPs in Shanghai.
First, spatial investment should prioritize high-frequency positive topics and reinforce distinctive place characteristics. Historic buildings, art exhibitions, photography, and social media check-ins were prominent in positive comments, while arts-oriented and leisure-and-entertainment CIPs generally achieved higher sentiment scores. Regeneration projects should therefore emphasize historic architecture, exhibition spaces, and landmark public spaces along major frontages and circulation routes. Careful design of lighting, materials, and sightlines can enhance visual legibility and photographic appeal, while architecture, exhibitions, activities, and rest areas should be integrated along continuous and legible visitor routes. Recurring programs, including exhibition seasons, themed markets, and evening openings, can sustain visitor interest and reduce reliance on isolated landmarks that may generate short-term attention but contribute little to long-term spatial distinctiveness.
Second, high-frequency negative topics should be used to identify minimum service and spatial standards for a reliable visitor experience. Complaints about sanitation, service, and crowding suggest that deficiencies in basic services and management are closely associated with negative overall evaluations. Priority measures include improved entrance legibility and wayfinding, stronger last-mile transport connections, and more effective crowd management during peak periods. CIPs should also ensure adequate sanitation facilities at key locations, monitor the spatial distribution of complaints, and evaluate the effectiveness of corrective actions. Standardized service procedures, capacity controls, queue management, segregated circulation where necessary, and emergency-response plans can further reduce variability in visitor experience and limit the recurrence of negative evaluations.
Finally, CIPs should establish a regular evaluation and feedback cycle linking spatial regeneration with day-to-day operations. Monthly mean sentiment scores and the proportion of negative comments can serve as monitoring indicators, while recurring negative topics can help identify specific operational or spatial problems. A park-level dashboard can track identified problems, corresponding corrective actions, and subsequent changes in public feedback. The persistence and diversity of positive topics can inform assessments of cultural and activity programming, while declines in recurring negative themes can be incorporated into operational performance evaluation. This process links changes in digital public perceptions to specific management responses and supports the continuous improvement of spatial quality and service provision
[42].
5.3 Strengths and Limitations
By analyzing a large UGC dataset, this study broadened the empirical coverage of public perceptions and identified fine-grained patterns in visitor evaluations. It developed an analytical framework that integrates sentiment analysis and topic modeling, complementing questionnaire- and interview-based approaches that are often constrained by sample size, temporal coverage, and self-reporting bias. The framework also enabled the examination of temporal variation in public perceptions and sentiment. By linking digital presence of place with reported experiences of physical space, the study provides insight into how social media and on-site experience interact in shaping perceptions of urban cultural spaces, and offers a replicable approach to analyzing public perceptions and an evidence base that can inform the spatial planning and management of CIPs.
This study also has several limitations that should be considered when interpreting the findings. Dianping comments are more likely to be generated by young and middle-aged users and consumption-oriented visitors, potentially underrepresenting nearby residents, employees, and visitors who do not engage in consumption activities. The platform’s rating categories and tags may encourage users to comment more explicitly on transport, service, and sanitation, while less tangible dimensions of place, such as local memory, cultural narratives, and social networks, may be less visible in the comment data. Reliance on a single platform also limits the ability to compare how digital presence circulates across platforms and how online attention is associated with subsequent visitation. Future research should integrate data from multiple platforms to assess platform-specific biases and combine social media data with questionnaires and in-depth interviews to capture sociodemographic characteristics and visit motivations
[43]. Such an approach would strengthen the interpretation of public perceptions and provide a more comprehensive understanding of how interactions between digital media and physical space can inform spatial planning and management.
6 Conclusions
Social media has extended public perception of CIPs from on-site experience to online expression. UGC now provides an important link between urban cultural spaces and public sentiment. Using Dianping comments of 37 representative CIPs in Shanghai, this study develops a framework of digital presence assessment, sentiment analysis, and topic identification. It combines CNN-based Senta sentiment analysis with Transformer-based BERTopic topic modeling to identify sentiment patterns and topic structures and to propose strategies for spatial regeneration and park management. The main findings are as follows:
1) CIPs show clear differences in digital presence. Central-city CIPs generally accumulated more comments and achieved greater platform visibility. However, some recently opened CIPs also attracted substantial online attention. These patterns suggest that UGC circulation can extend a CIP’s visibility beyond geographically proximate audiences.
2) Public sentiment is predominantly positive, with clear differences across CIP types. Approximately 85% of comments were classified as positive, with arts-oriented and leisure-and-entertainment CIPs exhibiting relatively higher sentiment scores. Positive evaluations were more closely associated with spaces that support participation, extended stays, and social media sharing. The findings further suggest a reinforcing sequence of “visit, comment, and sharing” may contribute to the accumulation of positive online attention.
3) The topic analysis revealed a multidimensional structure in which both distinctive spatial attractions and basic service conditions are associated with public perceptions of the CIP. Children recreation, transport accessibility, and historic buildings are among the most frequently discussed topics. Historic architectural character, art exhibitions, photography, and social media check-ins are prominent topics in positive comments, whereas sanitation, service quality, and crowd management are more prevalent in negative comments. CIP reputation therefore reflects both the appeal of shareable spatial features and the reliability of basic services and management.
Based on these findings, this study proposed an evidence-based approach to spatial optimization strategies. High-frequency positive topics can guide investment in distinctive and shareable spatial features, including historic architecture, exhibition spaces, and recurring cultural events, thereby reinforcing spatial identity and place character. Recurring negative topics can help identify priorities for improving transport connectivity, sanitation facilities, service quality, and crowd management. UGC-based sentiment and topic indicators can also support a continuous monitoring and feedback cycle linking spatial regeneration with CIP operations and public evaluation. Together, these measures provide a data-informed basis for the adaptive regeneration and management of CIPs.
PDF
(2744KB)
