Public Perceptions of Ecology, Facilities, and Social Atmosphere in Bangkok’s Urban Parks: An Exploratory Analysis of 7,585 Google Maps Reviews

1. Introduction

Urban parks support recreation, restoration, social interaction, biodiversity, and multiple cultural ecosystem services. Conventional park evaluation commonly relies on surveys, site audits, or structured observation. These methods remain important, yet the growth of user-generated content has created a complementary source of evidence: unsolicited descriptions of actual park experiences. Online review text can reveal what visitors notice, value, praise, and criticize at scales that are difficult to obtain through conventional fieldwork alone [1,2,3,4].

Recent studies have used social-media and online-review data to examine cultural ecosystem services, accessibility, amenities, landscape perception, and visitor satisfaction [5,6,7,8,9]. The resulting evidence suggests that public experience is multidimensional. Natural features such as trees, shade, water, and greenery can shape restorative and aesthetic perceptions, while facilities, maintenance, paths, toilets, playgrounds, and activity infrastructure influence practical use. Social conditions—including perceived busyness, quietness, noise, relaxation, and the presence of other people—also shape the atmosphere of public space.

Bangkok is a useful setting for this type of analysis. Its urban parks range from major metropolitan green spaces to smaller neighborhood and specialized parks, and they operate within a dense tropical city where heat, air quality, recreation, mobility, and access to nature intersect. Prior Bangkok-focused work has shown the value of digitally enabled analysis for understanding urban park experience [10]. However, comparative evidence across a wider set of Bangkok parks remains limited.

This study develops a descriptive baseline from a large archive of Google Maps reviews covering 20 Bangkok parks. The analysis addresses three practical questions: (1) how frequently reviews mention ecology-related features; (2) how frequently they mention facilities and activity infrastructure; and (3) how frequently they contain language related to social atmosphere, including crowding, busyness, quietness, peace, relaxation, and noise. A further objective is to identify how these signals vary across parks. The emphasis is therefore on transparent comparison of recurring themes in visitor narratives, providing a foundation for subsequent validation and more context-sensitive text analysis.

2. Materials and Methods

2.1. Study Design and Park Selection

The study used a purposive sample of 20 urban parks and park-like public green spaces in Bangkok. The project database records several complementary selection rationales, including variation in review volume, inclusion of major flagship parks, medium district parks, smaller neighborhood parks, geographic spread across Bangkok, and variation in ecological, facility, and social profiles. The sample was assembled to capture a broad range of urban park contexts rather than to estimate the prevalence of characteristics across all parks in Bangkok.

The selected parks differ markedly in scale, function, and review intensity, allowing the analysis to compare visitor narratives across heterogeneous urban green spaces. The set includes high-profile metropolitan destinations such as Lumphini Park and Benjakitti Park as well as district, neighborhood, memorial, sports-oriented, and specialized green spaces. This diversity is important for descriptive comparison because the prominence of ecology, facilities, and social atmosphere may vary with the way a park is designed, used, and experienced.

2.2. Review Corpus and Text Preparation

The project archive contains an initial collection of 14,965 Google Maps review records. After filtering for records with non-missing English review text, the analytical dataset comprised 7,585 reviews distributed across the 20 study parks. Multilingual review text had been translated into English within the project workflow before analysis. English text was converted to lowercase to standardize matching across the corpus.

The review was the unit of analysis. Reviews were retained even when none of the focal lexicon terms occurred so that theme incidence could be calculated against the full translated textual corpus rather than only against reviews already containing a target term. Park identifiers were used to link individual reviews to the corresponding park summary, enabling consistent aggregation of review-level signals to the park level.

2.3. Exploratory Theme Signals

Three lexicons were defined to capture recurring dimensions of park experience. The ecology lexicon contains the terms tree, nature, shade, green, bird, flower, lake, lizard, water, forest, and animal. The facility lexicon contains bathroom, toilet, track, run, walk, clean, dirty, parking, playground, and gym. The third lexicon was originally labeled ‘Crowd’ and contains crowd, busy, quiet, peace, people, relax, hustle, bustle, noise, and calm. Because this set combines congestion-related terms with words associated with tranquility and restoration, it is described here as a social-atmosphere signal rather than as a direct measure of crowding.

The ecology terminology was informed by broad ecological concepts [11], facility-related terms by literature on park activity and built-environment observation [12], and the social-atmosphere terms by work on soundscape and perceptual accessibility [13,14]. Occurrences were counted through literal string matching in the lowercased review text. A review received a value of 1 for a theme when at least one term from the corresponding lexicon was present and 0 otherwise. Because the three themes were coded independently, a single review could contribute to more than one theme.

2.4. Descriptive Analysis and Interpretation

For each theme, the analysis first identified whether a review contained at least one lexicon term and then calculated the proportion of reviews with that signal. The same procedure was applied at the park level, where incidence was calculated as the number of reviews containing at least one term from a given lexicon divided by the total number of translated textual reviews available for that park. These percentages provide a direct and interpretable measure of how often each theme appears in visitor narratives.

The project script also fits binomial generalized linear models to keyword-count features and generates theme probabilities. In the present analysis, the direct binary incidence measures were used for reporting because they preserve a straightforward relationship between the observed text and the park-level summaries. Environmental fields, including temperature, humidity, PM2.5, and PM10, are retained in the park summary file, but they were not included in inferential analysis because the available environmental observations are not temporally aligned with the historical review texts.

3. Results

3.1. Corpus-Level Theme Incidence

The analytical corpus contained 7,585 translated textual reviews. Facility-related language was the most common of the three focal signals, appearing in 2,731 reviews (36.0%). Social-atmosphere language appeared in 2,237 reviews (29.5%), while ecology-related language appeared in 1,955 reviews (25.8%). In total, 4,480 reviews (59.1%) contained at least one of the three focal signals. These proportions describe mention incidence, not positive or negative sentiment.

Table 1. Corpus-level incidence of the three exploratory review signals.

Signal	Reviews with signal (n)	Total reviews (n)	Incidence (%)
Ecology	1955	7585	25.8
Facilities	2731	7585	36.0
Social atmosphere	2237	7585	29.5
At least one focal signal	4480	7585	59.1

Table 1. Corpus-level incidence of the three exploratory review signals.

Signal	Reviews with signal (n)	Total reviews (n)	Incidence (%)
Ecology	1955	7585	25.8
Facilities	2731	7585	36.0
Social atmosphere	2237	7585	29.5
At least one focal signal	4480	7585	59.1

3.2. Variation Across Parks

Theme incidence varied substantially across parks (Table 2). Ecology-related language ranged from 6.0% at Nong Chok Public Park to 45.0% at Lumphini Park. Facility-related language ranged from 14.4% at Rak Thale Bangkhunthian Bridge to 50.1% at Benjakitti Park. The social-atmosphere signal ranged from 17.2% at Nong Chok Public Park to 54.2% at Charan Phirom Park; however, the latter estimate is based on only 24 reviews and should not be interpreted as a stable park-level prevalence estimate.

Among parks with at least 100 reviews, Lumphini Park showed the highest ecology signal (45.0%), Benjakitti Park the highest facility signal (50.1%), and Princess Mother Memorial Park the highest social-atmosphere signal (40.3%). Several large parks showed mixed profiles rather than dominance by a single theme. For example, Chatuchak Park combined relatively high ecology (32.3%), facility (43.5%), and social-atmosphere (36.9%) incidence, while Queen Sirikit Park combined comparatively high ecology (35.4%) with moderate facility (34.8%) and social-atmosphere (32.3%) incidence.

3.3. Interpretation of the Descriptive Profiles

The facility signal was the most prevalent overall, suggesting that visitors frequently use online reviews to describe the functional and operational qualities of parks—walking and running, tracks, toilets, cleanliness, parking, playgrounds, and gyms. This pattern is consistent with research showing that facilities and maintenance can strongly shape perceived park quality and satisfaction [7,8]. At the same time, the prominence of ecology-related language in parks such as Lumphini, Queen Sirikit, Chatuchak, and Santiphap indicates that visible nature remains a salient component of urban park experience.

The social-atmosphere signal requires more cautious interpretation. Because the current lexicon combines words associated with density or disturbance (such as crowd, busy, hustle, bustle, and noise) with restorative terms (quiet, peace, relax, and calm), a high percentage indicates that social or ambient atmosphere is frequently discussed; it does not indicate that a park is necessarily perceived as overcrowded. This distinction is important for later validation and is one reason the present paper avoids labeling the third signal as a direct crowding score.

4. Discussion

The analysis shows that Google Maps reviews provide a useful comparative lens on Bangkok’s urban parks. Across the 20 sites, facility-related language was most common, followed by social-atmosphere and ecology signals. This pattern suggests that online park reviews function not only as descriptions of natural qualities but also as practical accounts of use, amenity, maintenance, accessibility, and the ambient conditions that shape everyday park experience.

The results align with a broader literature in which online reviews have been used to identify context-specific determinants of park satisfaction, accessibility, and cultural ecosystem services [3,4,5,6,7,8,9,15]. For managers, this type of analysis can function as an agenda-setting tool. Repeated facility mentions can point toward the salience of paths, toilets, cleanliness, parking, playgrounds, and exercise infrastructure. Ecology mentions can reveal where trees, shade, water, flowers, birds, and other natural features are especially visible in visitor narratives. Social-atmosphere language can identify parks where quietness, relaxation, busyness, people, or noise are central to the visitor experience.

The approach should complement rather than replace surveys, field audits, or human interpretation. Online reviewers are self-selected and are not a probability sample of park users. The same percentage can also arise from very different meanings. For example, the word ‘clean’ may be positive, while ‘dirty’ is negative; both currently contribute to the same facility signal. Likewise, ‘quiet’ and ‘crowd’ both contribute to the social-atmosphere signal. A management system built from these data should therefore separate topic presence from evaluative direction before making operational judgments.

The present results also support a staged research strategy. A first stage can use transparent lexical screening to organize a large corpus and identify candidate differences. A second stage should use human coding to validate theme definitions and distinguish positive, negative, and neutral evaluations. A third stage can test more advanced NLP or large-language-model classification against those human labels. This sequence would preserve interpretability while allowing stronger modeling later.

4.1. Limitations

Several limitations define the appropriate interpretation of the results. First, Google Maps reviews reflect voluntary platform participation and therefore contain selection and platform biases. Second, multilingual reviews were translated into English before analysis; translation can alter wording, nuance, and culturally specific expressions. Third, the current matching procedure uses literal string counts rather than tokenized or context-aware semantic classification, so terms may be affected by polysemy, negation, morphology, and substring matching. Fourth, the three lexicons measure mention incidence rather than sentiment, importance, or causal influence.

The social-atmosphere lexicon also combines crowding-related and restorative terms, which limits its interpretation as a single construct. Human-coded validation of the three themes has not yet been incorporated into the analysis, and some parks have relatively small review counts, making extreme park-level percentages less stable. In addition, the park sample was purposively constructed for diversity and is not representative of all Bangkok green spaces. Finally, the environmental fields retained in the project database are not temporally matched to individual review dates; accordingly, this study does not test relationships between historical visitor perceptions and current PM2.5, temperature, or humidity values.

5. Conclusions

A corpus of 7,585 translated Google Maps reviews from 20 Bangkok parks shows clear differences in what visitors discuss. Facility-related terms appeared in 36.0% of reviews, social-atmosphere terms in 29.5%, and ecology-related terms in 25.8%, with 59.1% of reviews containing at least one focal signal. The park-level profiles varied substantially, with Lumphini Park standing out for ecology mentions, Benjakitti Park for facility mentions, and several parks showing distinctive social-atmosphere profiles.

The principal value of the analysis lies in its methodological transparency and comparative coverage. Review analytics can organize large volumes of user-generated content and reveal where nature, facilities, and social atmosphere are especially salient in public narratives. Further human coding and context-sensitive analysis can strengthen interpretation by distinguishing topic presence from evaluative direction and by testing whether the lexicon-based patterns remain stable under more rigorous thematic classification. Together, the results provide a reproducible descriptive baseline for continued research on public perceptions of Bangkok’s urban parks.