The sustainable management of overtourism via user content

The sustainable management of overtourism via user content ☆ Concepci´ on Foronda-Robles a , Luis Galindo-P´ erez-de-Azpillaga b,* , Pablo Armario-P´ erez c a Instituto Andaluz de Investigaci´ on e Innovaci´ on en Turismo (IATUR), Universidad de Sevilla, Av/ San Francisco Javier, S/N, 41018 Seville, Spain b Departamento de Geografía Humana, Universidad de Sevilla, Instituto Andaluz de Investigaci´ on e Innovaci´ on en Turismo (IATUR), C/ María de Padilla S/N, 41004 Seville, Spain c Instituto Andaluz de Investigaci´ on e Innovaci´ on en Turismo (IATUR), Universidad de M´ alaga, Blvr. Louis Pasteur, 29, Jardín bot´ anico Campus Teatinos, 29071 Malaga, Spain ARTICLE INFO Editor: Dr. Ksenia Kirillova Keywords: UGC analytics Online reviews Sentiment Tourism impacts Spatial analysis Adaptive responses ABSTRACT This research evaluates impacts of overtourism in Granada, Spain, by analysing 1349 negative comments from TripAdvisor for 71 tourist attractions. Employing a mixed-methods approach, the study uses sentiment analysis via the BERT model, multivariate analysis (PCA and K-means clustering) and social network analysis. Key findings reveal issues of congestion, high costs and environmental degradation, identifying user satisfaction and spatial significance as critical dimensions. The study highlights a strong positive correlation between AI-driven sentiment and user opinions. Practical implications underscore the need for sustainable management strategies—including destination diversification, improved transport networks and access control—to mitigate highly touristified environments, preserve visitor experience and protect local heritage, thereby promoting sustainable tourism. Online reviews are deemed valuable for proactively addressing tourist concerns. 1. Introduction Tourist perception is decisive for destination satisfaction and is intrinsically linked to sustainability. The relationship between perception and satisfaction is complex (Papadopoulou et al., 2023), as various aspects of the destination influence visitors’ experiences (Ramesh & Jaunky, 2021). Musa et al. (2024) highlight the role of sustainability in enhancing tourist experiences, noting that tourists prioritise ethical and environmental concerns over cost. User-generated content (UGC) has become a valuable source of information on tourist perceptions (Cheung et al., 2022; Wang et al., 2021; Yan & Halpenny, 2023), providing realtime comments on satisfaction and concerns. Barbosa (2022) defines UGC as “any kind of text, data or action performed by online digital systems users, published and disseminated by the same user through independent channels, that incur an expressive or communicative effect”. Overtourism, a challenge of tourism growth, exacerbates sustainability issues by exceeding destinations’ carrying capacities (Butler & Dodds, 2022; Pechlaner et al., 2024; Santos-Rojo et al., 2023). Pham et al. (2024) state that tourism surpasses the social and environmental capacities of communities, threatening environmental balance, economic sustainability and socio-cultural cohesion (Mihalic, 2020). Therefore, it is essential to analyse how overtourism affects tourist satisfaction and its implications for sustainable tourism management. Economic factors, such as the price and perceived value of tourism services, fundamentally shape satisfaction (Huang & Chelliah, 2024). Tourists assess costs relative to service quality, this influencing their sustainability perception (Soltani-Nejad et al., 2024). Overtourism raises prices, limiting access for residents and tourists, and generating dissatisfaction (Moreno-Izquierdo et al., 2023). Environmental impacts perceptions in crowded areas strain ecosystems (Tkaczynski et al., 2022), exceed carrying capacities and diminish natural appeal (Rempel et al., 2021; Tokarchuk et al., 2022). Sociocultural factors also play a role: cultural authenticity and positive resident interactions boost satisfaction (Genc & Gulertekin Genc, 2023; Ramkissoon, 2023), while hostility or perceived invasion can lead to tourismphobia (Milano et al., 2024), causing friction between visitors and hosts (Chien et al., 2024). Positive emotions correlate with higher satisfaction (Nieves-Pav´ on et al., 2024; Yan & Halpenny, 2023), whereas negative emotions correlate with a lower level (Shah et al., 2024; Sultana et al., 2023; Yin et al., 2023). In the digital age, online reviews aid understanding of tourists’ perceptions and satisfaction (Guo & Pesonen, 2022). Platforms like ☆ This article is part of a Special issue entitled: ‘Monitoring and evaluating’ published in Annals of Tourism Research Empirical Insights. * Corresponding author. E-mail addresses: [email protected] (C. Foronda-Robles), [email protected] (L. Galindo-P´ erez-de-Azpillaga), [email protected] (P. Armario-P´ erez). Contents lists available at ScienceDirect Annals of Tourism Research Empirical Insights journal homepage: www.sciencedirect.com/journal/annals-of-tourism-research-empirical-insights https://doi.org/10.1016/j.annale.2025.100184 Received 28 December 2024; Received in revised form 16 June 2025; Accepted 18 June 2025 Annals of Tourism Research Empirical Insights 6 (2025) 100184 Available online 28 June 2025 2666-9579/© 2025 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/bync-nd/4.0/ ). TripAdvisor reflect tourists’ experiences and indicate perceived sustainability (Chang & Ko, 2024; Zheng et al., 2023). These reviews provide insights that help to address dissatisfaction related to economic costs, environmental degradation and social congestion. This article explores how overtourism influences the tourist experience and its expression in online reviews of Granada’s attractions. The study evaluates user satisfaction and its spatial significance using the BERT model (Bidirectional Encoder Representations from Transformers), AI indicators and multivariate analysis. It offers a novel perspective on digital feedback and proposes strategies for harmonising growth with sustainability in destination management. 2. Literature review 2.1. Overtourism In the 1960s, tourism studies highlighted its positive impacts, such as economic development, infrastructure improvement and cultural exchange. However, from the 1970s, the negative effects on natural and heritage resources became evident, prompting a shift in research towards tourism sustainability. Overtourism, characterised by the negative impacts of excessive tourist influx, has been analysed from various perspectives, including impact perception and negative experiences (Goodwin, 2021), excessive carrying capacity (Pechlaner et al., 2024) and adverse socioeconomic consequences (Coconi et al., 2024). Overtourism reflects the negative perception of tourism costs exceeding its benefits (De Miguel et al., 2023). Gowreesunkar and Thanh (2020) note that overtourism deteriorates infrastructure, raises living costs and generates social tensions. Sustainable management strategies are needed to balance economic benefits with resource protection and tourist experience (Buitrago-Esquinas et al., 2023; Dilshan & Nakabasami, 2025). 2.2. User-generated content (UGC) in Tourism UGC refers to spontaneous and authentic feedback shared by consumers on digital platforms (Guo et al., 2017). This content provides valuable insights into tourist experiences, aiding in perception analysis and strategy design based on real user experiences (Yan et al., 2024). Online reviews reflect visitors’ satisfaction and disappointment (Garner & Kim, 2022), capturing their experiences and perceptions at destinations (Van der Zee et al., 2020), whether examined emotionally or rationally (Cheung et al., 2022). Supported by customer value theory, UGC helps monitor destination image and reputation, analysing both positive and negative sentiments to identify strengths and weaknesses (Bigne et al., 2020). Recent studies on this issue have proliferated, analysing online reviews on platforms such as TripAdvisor, Lonely Planet and Google Maps (Cheng et al., 2019). While many studies focus on hotels and restaurants (Cenni, 2024; Orea-Giner et al., 2022), fewer address tourist attractions (Ali et al., 2021; Cherapanukorn & Sugunnasil, 2022; Guerrero-Rodriguez et al., 2023), which have contextual dimensions (Taecharungroj & Mathayomchan, 2019) unlike the functional attributes of hotels and restaurants. Tourist perceptions are situational (Butler & Dodds, 2022), influenced by local residents’ attitudes (Oklevik et al., 2019), weather patterns, and seasonality (Dr˘ agan & Camar˘ a, 2022), all of which affect enjoyment and planning. High visitor concentrations at attractions (Mckinsey and WTTC (World Travel and Tourism Council), 2017) can lead to dissatisfaction due to noise or resource access issues. As for the reviews, both their usefulness and the scores contained therein should be interpreted with caution (Ali et al., 2021). Kim and Hwang (2020) examine how the relationship between the review ratings and the perceived use of these varies. Their analysis reveals the existence of a bias towards negativity in the expression of emotions: negative emotions prove to be more useful in reviews with positive scores than in those with negative scores. Another point to consider, according to Ransbotham et al. (2019) is that smartphone reviews tend to be more effective than the other reviews, since they are written immediately after the experience. In numerous studies of online reviews, the authors tend to pay more attention to negative comments (Lee et al., 2017). Then, reviews written by foreign tourists on the quality of experiences usually come with a higher rating (Ganzaroli et al., 2021). Cuadros-Mu˜ noz (2024) examines various apps that analyse feelings and emotions, and remarks that many of these tools display significant limitations. This author highlights that some of the free versions of these apps have very restricted functionalities or are designed to only process text in English. The spatial distribution of opinions on a destination on TripAdvisor reveals the concentration of visitors at ‘hot spots’, visited more frequently, and ‘cold spots’, visited less frequently (Van der Zee et al., 2020), which might also be influenced by tourist flows (Foronda-Robles et al., 2022). These geographical patterns are useful for understanding a destination’s degree of saturation, providing valuable information on both online and offline tourist behaviour. This information is key for a destination’s managers to be able to identify and analyse those areas most affected by complaints, understand the sources of the observed negativity and take measures to improve the tourist experience. Yu and Egger (2021), in their study on Paris, highlight how a feeling of anxiety can generate uncertainty in tourists due to a high concentration of people. They also point out that overtourism adversely affects tourism services, especially with respect to waiting times, a problem they suggest could be solved with technology. Studies on overtourism are numerous, yet there is a significant gap in the analysis of tourist perceptions through UGC. While digital platforms and social networks provide valuable insights into tourist experiences, few studies have systematised the data they provide to evaluate the effects of overtourism (Butler, 2020). This approach enables inductive analysis of tourists’ perceptions at saturated attractions, using text analysis tools to identify key issues related to overtourism (Koens et al., 2018; Mellinas et al., 2024; Yu & Egger, 2021). This gap offers an opportunity to explore problems and design more effective management strategies. 3. Method 3.1. Participants TripAdvisor is one of the most influential platforms in the tourism sector, providing an excellent collection of data for tourism research. This portal allows users to compile information on their trips, evaluate them and share comments, and is a pioneer in the adoption of UGC models. The platform’s services are free for consumers, who provide a significant proportion of the content. However, there have been questions asked about the motivations behind the opinions given (Ali et al., 2021; Filieri et al., 2021; Roy et al., 2024). The city of Granada, located in Andalusia in the south of Spain, is a tourist destination that offers diverse attractions and experiences. In 2023, Granada recorded 3.15 million overnight stays, with 51 % from foreign tourists, mainly from the USA, France, Italy and Germany (NSI - National Statistics Institute of Spain, 2024). These tourists impact the city, altering the functionality of uses and adapting the offer to the fluctuating population. Cerezo-Medina et al. (2022) warn of the potential effects of the exponential growth of tourist accommodation and the residents’ feelings of rejection towards tourism. The Alhambra Palace and the Generalife Gardens therein are the most popular attractions, leading multiple lists and rankings, such as those Tripadvisor (2024) and Spanish Tourism (2024). Both monuments are leaders in Spain, with 2.6 million visits (Patronato de la Alhambra y Generalife, 2023). Various publications have highlighted the influence of these two references on the distinctive perception of Granada as a brand on social media networks (Peco-Torres et al., 2021), as well as their strengths and weaknesses based on visitors’ opinions (Prados-Pe˜ na C. Foronda-Robles et al. Annals of Tourism Research Empirical Insights 6 (2025) 100184 2 et al., 2021). The selection of Granada as a tourist destination may be skewed if visitors’ interpretations of popular attractions predominate over the city as a whole. Valverde-Roda et al. (2023) emphasise the importance of conducting interviews across various parts of the city, not only in the most visited areas. Such studies must take a balanced view and consider all the territorial diversity, viewing destinations as part of a broader system, especially in overtourism research (Fyall & Garrod, 2020). It is also important to restrict general interpretations associated with the prevalence of the visual aspect, as this component tends to be given excessive importance in online comments, eclipsing other equally significant aspects (Alvarado-Sizzo, 2023). The fascination with capturing and sharing images has led to a search for tourist attractions at times being reduced to a mere visual effect. This minimises other types of information that tourists can provide on digital platforms such as TripAdvisor, in particular on the role they play in the whole of which they are a part, or detracts from other tourism attractions that coexist in the same territory (Karayazi et al., 2021). Along the lines of Zhao et al. (2019), it is proposed that extreme cases be excluded from the analysis in order to avoid an over-representation essentially based on these visual aspects, which could hinder a full understanding of the global context of the attractions that make up the city of Granada. For this reason, the sample does not include the Nasrid Palaces (the Alhambra Palace and the Generalife Gardens), the Cartuja Monastery and the San Nicolas Viewpoint. After excluding these attractions, there remain comments relating to a total of 232 existing tourist attractions (this number can vary depending on when exactly the data was mined) taken from the “Things to do in Granada” section of TripAdvisor. 3.2. Instruments Table 1 presents two indicators used to measure the evaluation of the tourist attractions based on the comments made. Each indicator provides a quantitative scale to evaluate interest, and together they offer an integral vision of the evaluation of these attractions, combining the artificial intelligence (AI) perspective and that of the users’ opinions themselves. Both variables, User_AI and User_Opinion, reflect the degree of satisfaction with the tourist destination. In the case of User_AI, the BERT (Bidirectional Encoder Representations from Transformers) model is used. There are studies that analyse tourism in Granada through the analysis of feelings, such as that carried out by Vi˜ n´ an-Lude˜ na and de Campos (2022), using the BERT model on data from X and Instagram. In this indicator, each comment is assigned an integral value between 0 and 4, both included, that represents the emotional evaluation. In addition, the model provides the probability that each comment will be classified by each value from p0 to p4. In order to associate the value of the feeling with any value in the interval [2–10], the following formula is used: (p0*0 +p1*1 +p2*2 +p3*3 +p4*4)*2 +2 This variable applies to all comments, regardless of the language, as the BERT model is pre-trained to avoid translation ambiguities in comments. For this purpose, two English corpora are used—BookCorpus and English Wikipedia (Bahri & Suadaa, 2023). Different hyperparameters and training series were used to pretrain the BERT model, including learning rate, batch size, epoch and weight decay. This model was implemented in 25 ways, with five models analysing the tokeniser’s function, which was limited to a vocabulary of 109,999 tokens. Repeating the same training process produced slightly different classifications, so another five models used the same hyperparameters to account for randomness. Of the remaining 15 models, the best classifier was trained with 350 manually classified comments. This model (Model ID: 12) learns in a controlled manner (learning rate =0.00002), processes one datum at a time (batch size =1) and repeats training five times (epoch =5), with regularisation (weight_decay =0.01) to avoid overfitting. When validated with 50 manually labelled comments, this model achieved the highest accuracy (96.30 %) in distinguishing between positive and negative comments. The classification error rate for both positive and negative comments was 26.32 %, outperforming the other models. In the second variable, User_opinion, the evaluations available on TripAdvisor are integrated. The users qualify the tourist attractions in their comments using a scale that assigns 10 to excellent, 8 to very good, 6 to average, 4 to poor and 2 to terrible. These ratings reflect the users’ perceptions of the tourism resource and provide a quantitative measurement of their satisfaction. Although these appraisals are subjective and can vary according to the individual expectations, previous experiences and personal preferences of each user, by compiling and analysing a large number of these evaluations, it is possible to secure a sufficiently impartial overview of the users’ satisfaction with regard to a specific tourist destination (Krey et al., 2024). 3.3. Data collection The data for User_AI and User_opinion were obtained through web scraping techniques, using tools such as Python and Apache Airflow, and the APIs of open data services. Web scraping for comments occurred on five weekdays in May and June 2023, during low activity periods on the target website, typically at night or in the early morning. The Python Scrapy framework created spider diagrams to extract information after preprocessing. Complementary libraries, notably Selenium, were used for executing JavaScript code. Apache Airflow coordinated the spider diagrams, distributing tasks among available machines. Data was stored in CSV format for later use (Barbera et al., 2023). During the comment extraction process, quantitative data were recorded, such as the date of the review, the numerical score awarded by the user and the number of comments. In addition, geolocalisation data were obtained for each of the tourist attractions, using latitude and longitude values, through web scraping techniques. The representation of the results was achieved using the Granada municipality layer, obtained from the Instituto de Estadística y Cartografía de Andalucía—IECA—(Andalusian Institute of Statistics and Cartography), which offers a statistical and geospatial data portal (DERA). Then, an urban road layer provided by the Instituto Geogr´ afico Nacional—IGN— (National Geographic Institute) was added for a more complete characterisation of the territory. A collection of 42,778 comments covering all the tourist attractions in the city (232 at the time of extraction) was made from TripAdvisor’s “Things to do in Granada” list, excluding over-represented attractions. During processing, tokenisation divided sentences into segments (Mao et al., 2022). Using the Google Wordpiece algorithm (Ravichandiran, 2021), all letters were converted to lower case to normalise the text and avoid redundancies (Egger & Gokce, 2022), followed by the removal of stop words, such as connectors and articles (Hou et al., 2019). Comments were refined to retain only relevant texts. In a subsequent review, comments that could not be processed automatically were included manually (Liu & Li, 2019). Table 1 Indicators for measuring the evaluation of tourist attractions. Designation Range Description References User_AI [2−10] Evaluation generated through the application of artificial intelligence (AI), using the BERT (Bidirectional Encoder Representations from Transformers) model. (Chu et al., 2022; G´ omez-Talal et al., 2024) User_Opinion [2–10] Evaluation provided directly by the user on the online platform. (Mirzaalian & Halpenny, 2021; Nowacki & Niezgoda, 2020) C. Foronda-Robles et al. Annals of Tourism Research Empirical Insights 6 (2025) 100184 3 3.4. Data analysis This study analyses both the accumulation value and the content of TripAdvisor comments, using a combination of quantitative and qualitative methods. After collecting and refining the comments, the analysis focused on terms related to overtourism, identifying patterns and trends (Dillette et al., 2021). The “Type” classification on TripAdvisor’s “Things to do in Granada” list was used—considering 20 categories, including neighbourhoods, historical places, museums and parks—to provide an overview of attractions. The analysis examined the relationship between polarity and intensity using Key Descriptors (KDs) (Chu et al., 2022). Three categories were defined: financial cost, environmental quality and congestion. A lexical corpus was developed, aligning with these categories (Yu & Egger, 2021). Using Power BI, 28 KDs were identified as frequently occurring and related to overtourism (Table 2). In each of these KDs, nouns and adjectives naturally combine in text analysis, as users express their experiences and opinions using both descriptive and evaluative terms. This overlap enriches category representation, capturing different nuances (Al-Ghuribi et al., 2020; Yu & Egger, 2021). Additionally, aspect-based sentiment analysis (ABSA) was used to determine the negative polarity of each KD. A systematic and contextualised compression of comments associated with the selected KDs was performed, avoiding automated modelling (Mellinas et al., 2024; Valdivia et al., 2017). Consequently, the frequency of negative evaluations (KDs) was reduced from 4374 to 1349 across 71 tourist attractions, ensuring all received either fully or partially negative assessments. Following this identification and delimitation of KDs, sequential analytical techniques were applied, integrating statistical, spatial and network analyses to provide a comprehensive and multidimensional perspective on tourism dynamics in the study area (Fig. 1). Following the sequence in the figure, the quantitative study involves adding observations to the scores each KD receives in the comments and defining three variables. Two variables are the average values of the KDs from the indicators: User_AI measures the sentiment associated with each KD, and User_opinion provides the mean user rating. The third variable, KD Tourist Attraction Count (KD-TAC), represents the total number of tourist attractions in Granada associated with each KD, indicating its territorial relevance. Once defined, the correlation coefficient measures the relationship between the three variables. Furthermore, a Social Network Analysis (SNA) of the KDs in each category is proposed. They are analysed with the Ucinet 6 (Version 6.742) software, which examines their frequency and context in each category and the relationship between them (Borgatti et al., 2024). Using a spreadsheet template, an adjacency matrix was created to represent the binary relationships between the variables and, subsequently, the Pajek package was obtained from the NetDraw program to graphically represent the structures of the relationships between the different nodes generated previously in the matrix analysis. Of the 1349 comments, 464 include more than 1 of the KDs, 410 including KDs in the Economic Cost category, 47 those in Environmental Degradation and 7 those in Social Congestion. A two-mode network is used (Fan et al., 2021). Once the SNA has been completed, a Principal Component Analysis (PCA) is applied, this being a method used to reduce the dimensionality of the data collected from TripAdvisor, preserving the largest amount of information possible and obtaining those components that express the greatest variability (Omuya et al., 2023). Then, a spatial analysis is undertaken using the QGIS 3.38 software to show the localisation of the tourist attractions in relation to a symbolically central element in the city. This enables an evaluation of the degree of dispersion or concentration of tourism supply and demand, as well as that of tourist accessibility and mobility. First of all, the tourist attractions are represented by geolocalised points and then the factor of distance from the centre is added. This allows the visualisation of two isochrones of 1000 and 2000 m from Puerta Real, the study’s point of reference (Campos-S´ anchez & Chill´ on, 2020). These isochrones are calculated using the Hqgis plugin, which measures the distance on foot. After incorporating these elements into the map of Granada, heat maps are produced for each attraction, weighted by the number-of-KDs variable. Finally, a K-means cluster analysis is carried out, a technique that groups KDs, according to the calculation of the data centroid, into different K groups to generate clusters of the KDs included in the users’ comments based on their similarities (H¨ opken et al., 2024). 4. Results From the 1349 observations that contain negative evaluations, Table 3 has been generated, providing the Pearson correlation matrix for the three variables described. The User_AI and User_opinion variables present a correlation coefficient of 0.831, which indicates a strong positive correlation. This demonstrates that the BERT model result is significantly influenced by users’ opinions on the platform. As the mean BERT model score rises for each KD, so too does the mean evaluation of users on TripAdvisor. This suggests that the BERT model selected efficiently captures the users’ general opinion. However, the relationship between User_AI and KDTAC is also positive, albeit with a more moderate correlation, specifically, a coefficient of 0.434 between the two variables. This in turn shows that when the mean score of the BERT model rises, so too does the number of tourist attractions associated with each key word. Thus, those KDs linked to a greater number of tourism resources will tend to receive more positive scores in their comments. Fig. 2 shows the SNA analysis that visualises the interconnection between the nodes corresponding to Granada’s tourist attractions (Node 1) and the KDs, that is, Economic Cost, Environmental Degradation and Social Congestion (Node 2). The nodes with greater centrality in the network represent those tourist attractions that are most connected to the issues identified (KDs), this indicating a greater impact on these places. The high density in certain parts of the network suggests a strong interconnection between the KDs for the most popular attractions. Nevertheless, in general, the density is moderate, which implies a wider distribution of the problems in Granada. The centralised attractions require priority actions due to high maintenance costs, environmental degradation and congestion. Measures are recommended, such as access limitation, the creation of alternative routes and the dispersion of the tourist flow to relieve the pressure. Table 4, within the PCA analysis, presents the squared cosines of the variables for three factors (F1, F2 and F3) that explain 100 % of the Table 2 KDs defined and their distribution by category. Categories KDs KDs in a review KDs in a negative review % of negative reviews Economic Cost (EconC) Quality, Expensive, Money, Free, Taxes, Fine, Price, Fee 3507 813 23.18 Environmental Degradation (EnviD) Abandoned, Waste, Decayed, Dilapidated, Neglected, Graffiti, Noisy, Dirty 456 321 70.39 Social Congestion (SociC) Crowd, Annoying, Queues, Stressful, Uncomfortable, Overcrowded, Irritating, Forbidden, Restricted, Saturated, Foreign/tourist 411 215 52.31 4374 1349 30.84 C. Foronda-Robles et al. Annals of Tourism Research Empirical Insights 6 (2025) 100184 4 variability in the data. These squared cosines indicate the quality of the the variables’ representation for each factor, based on the factor loads and the correlations between the variables and the factors. A value close to 1 suggests that the variable is well represented. The variables, User_AI and User_opinion, have a strong correlation to factor F1, with squared cosines of 0.900 and 0.792, respectively, which mark this factor as the dimension representative of user satisfaction or quality perception for the tourist on TripAdvisor. On the other hand, the KD-TAC variable has a notable correlation to F2, with a squared cosine of 0.633, marking this dimension as that of the spatial significance or impact on tourist sites configuration. In Fig. 3 can be seen the applied factor scores for the observations. F1 explains 68.503 % and F2 26.635 % of the variability, representing a total of 95.138 % between the two dimensions (Appendix 1 lists the squared cosines of each observation). These factor scores enable an interpretation of which aspects of the tourist experience are most strongly associated with each KD. In F1, which represents the measurement of user satisfaction, the KDs strongly associated with this dimension are key indicators in this area. The greatest positive contribution (squared cosine higher than 0.5 in the dimension) is the “expensive” KD (2.004). This relationship indicates that although the term “expensive” has a negative connotation, in the context of the comments, its negative impact is mitigated in the quality perception dimension. This in turn indicates that in spite of the high costs that certain tourist attractions may incur, visitors continue to evaluate their experience positively. Meanwhile, the greatest negative contribution is from the KD “fee” (−8.742), which is strong in nature but deviates from the general trend (not represented in Fig. 3). Of greater interest is the KD “taxes” (−3.093), where another economic aspect of high costs is associated with a significant reduction in the perception of satisfaction. This indicates that those terms related to additional costs, such as “taxes” and “fee”, tend to adversely affect tourist satisfaction, since they are perceived as payments made that are not directly associated with the tourism activity itself. The F2 axis represents the spatial significance of the KDs. In the city of Granada, there is a high concentration of tourist attractions in the centre. A total of 32 attractions (45.07 % of those included in the study) are located at a distance of 1 km from Puerta Real. By extending this distance to 2 km, which is a reasonable distance for a tourist to cover on foot, 61 attractions (85.92 %) are arrived at. As a general reference, 64.56 % of the KDs are associated with attractions located within these two defined isochrones (Fig. 4). Defined areas of influence: 1. Capilla Real de Granada-Plaza Bib Rambla, 2. Albayzin, 3. Calle Navas, 4. Carrera del Darro, 5. Sacromonte, 6. Parque de las Ciencias, 7. Plaza Nueva, 8. Calle Elvira, 9. Paseo de los Tristes, 10. Monasterio San Jeronimo, 11. Carmen de los Martires, 12. San Miguel Alto, 13. Museo Cuevas del Sacromonte. This factor (F2), related to the localisation in the city of tourist Fig. 1. Map the interconnections between KDs. Table 3 Correlation matrix (Pearson). Variables User_AI User_opinion KD-TAC User_AI 1 0.831 0.434 User_opinion 0.831 1 0.251 KD-TAC 0.434 0.251 1 The values in bold are different to 0, with an alpha significance level of 0.05. C. Foronda-Robles et al. Annals of Tourism Research Empirical Insights 6 (2025) 100184 5 attractions with comments, explains a significant variability, albeit lower than that of the previous factor. The greatest positive contribution comes from the KD “price” (0.722), which indicates a greater distribution of comments related to costs at tourist attractions in Granada. The most notable negative contribution is provided by the KD “crowd” (−2.564), which although it affects less attractions, generates negative comments and is located in neighbouring areas such as Albayzin and Sacromonte. In a joint interpretation of the two axes, the KDs with high scores in both F1 and F2 appear to be capturing aspects that converge in Granada’s tourist experience. These linked positive values are located in the upper right quadrant, with the KDs “expensive” and “price” standing out. In the same way, the KDs with low scores in both factors can also be associated. There are some significant observations with negative values in the bottom left quadrant, such as the abovementioned KDs “taxes” and “fine”, once again from the Economic Costs category. The proximity to or distance from the centre of the axes determines the intensity with which the KDs are used. To complete the multivariate analysis, the K-means cluster provides a valid interpretation of how the user comments are grouped, based on the three variables mentioned. The results enable the identification of patterns and trends in the comments for decision making and strategy formulation by the relevant bodies. Table 5 displays the results of the four clusters, including the centroid of each variable that represents the average value of the observations within each class. Then Fig. 5 completes the information, showing the silhouette scores of each KD. Each score is a measure of how similar an object is to its own cluster (cohesion) in comparison with other clusters (separation). Class 4 represents the KDs where users give a high evaluation and where the greatest number of attractions is referred to, but whose comments vary most significantly. This is reflected in an average centroid cluster score for the BERT model of 6.799, for direct opinion of 6.983 and for the total number of attractions of 32.857. This class is the smallest but the highest weight accumulation (470). However, its variance is also higher (2881.108) than that of the other classes, which indicates the great variability between the two objects. Class 4 confirms that the two objects are dispersed, the KDs being “expensive” and “price”. In general, these KDs represent comments that express concerns about the high prices of access to monuments and other attractions. They achieve the highest opinion scores and the greatest number of associated attractions, because of which they could be of particular interest to tourism management in Granada. These KDs mainly coincide with spaces that have a strong commercial component (Calle Navas, Plaza Nueva, Calle Elvira, Alcaiceria and Plaza Bib Rambla). Thus, greater investment in improving certain attractions, alongside establishing more affordable and competitive entrance fees at other attractions, could help to relieve congestion and decentralise some areas of the city by redirecting the flows. All the goods and services available in these areas, through restaurants, souvenir shops and access to monuments, for example, have inflated prices due to high tourist demand. Access to the monuments is especially expensive, which limits accessibility for some visitors and can generate a perception of economic exploitation. Classes 1 and 3 represent the KDs where users make the most consistent comments. Class 3 has an average score for the BERT model of 5.646, for direct opinion of 6.027 and for the total number of attractions of 13.293. This class includes 4 observations and a weight accumulation of 225. Its variance within the class (57.461) is similar to that of Class 1, which indicates a similarity in the KDs represented in the two classes. In the contrasts between variables, in relation to User_AI and User_opinion, the greatest difference appears precisely between the abovementioned Fig. 2. SNA applied to the tourist attractions and the KDs. Table 4 Squared cosines of the variables. F1 F2 F3 User_AI 0.900 0.022 0.078 User_opinion 0.792 0.144 0.063 KD-TAC 0.363 0.633 0.005 The value in bold for each variable corresponds to the factor for which the squared cosine is the greatest. C. Foronda-Robles et al. Annals of Tourism Research Empirical Insights 6 (2025) 100184 6 Class 4 and this Class 3, with 1.153 and 0.956 evaluation points, respectively. This signals a divergence of approximately 1 point in the appreciation of user opinion between both classes in the measurement of user satisfaction, because of which, if Class 4 is representing the most satisfied tourists, the least satisfied are in Class 3, according to the information gathered on the platform. The KDs of degradation are “noisy”, where there is an abundance of complaints about Calle Elvira, and likewise, “waste” and “dirty” about Fig. 3. Factor scores of the observations (axes F1 and F2 =95.138 %). Fig. 4. Areas of influence of tourist attractions with the most negative comments. C. Foronda-Robles et al. Annals of Tourism Research Empirical Insights 6 (2025) 100184 7 Albayzin. Also, the KD “free” stands out in the economic component. Class 3 thus represents opinions or comments that mostly express environmental concerns. Class 1 has 5 objects and a weight accumulation of 364. The KDs in the environmental category are “abandoned”, “neglected” and “dilapidated”. Carmen de los Martires, Ermita de San Miguel Alto and Palacio de los Olvidados are cultural attractions that could help to relieve congestion in the centre and provide an additional cultural offering, but they present certain challenges. Cost KDs that appear in this grouping are “quality” and “money”. This class has the lowest average distance from the centroid, which indicates that on average the objects are closer. Specifically, the KD “abandoned” has a silhouette score of 0.840, which shows that it is in the appropriate grouping. All the KDs in this class have high silhouette scores (close to 1), because of which they are near to other objects in their cluster and far from objects in other clusters. The KDs in this class represent comments that express concerns about the maintenance of a service or product, associated with the degradation from which they may be suffering. The fact that “money” and “quality” are among the KDs also confirms the existence of visitors’ concerns about price in relation to perceived quality. Last of all, Class 2 is in an intermediate position, where there is greater variability in the observations compared with Classes 1 and 3 but less than in Class 4. This is the largest grouping, with 16 objects and a weight accumulation of 290. Although it has largest number of objects, the variance within the cluster is moderate (124.111), which suggests a certain variability in the opinions represented by the KDs in the class. For the KD-TAC variable, the greatest range (24.091) is found precisely between Classes 4 and 2, which indicates that there is a contrast between these two clusters in the number of attractions affected, the latter accounting for the least spatial significance. The KDs in this class concerning congestion include “crowd” and “queues”, especially in relation to Paseo de los Tristes and Carrera del Darro—which, in spite of their great popularity among tourists, generate unease due to how narrow they are, the vehicles passing through and the high flow level of both tourists and local residents on foot—and Parque de las Ciencias when there is a high concentration of school visits. In the cost KDs is “taxes” and in the environmental KDs is “decayed”. This class thus represents a wider range of opinions, related to the users’ experience of comfort, efficiency and regulations. It can be seen that the tourists in this grouping often comment on crowds or a feeling of distress during visits, although it should be taken into account that some congestion KDs, such as “overcrowded”, have a silhouette score of −0.614, which means that this and other similar KDs might be close to the decision limit of their neighbouring clusters. 5. Discussion This study analyses the impact of overtourism on the tourist experience in Granada, using online reviews as the primary source. Key findings include: (1) issues related to economic cost, environmental degradation and social congestion; (2) fundamental dimensions of tourist perception, such as user satisfaction and the spatial significance of attractions; and (3) critical areas with multiple problems requiring specific management measures. These findings highlight the importance of online reviews for understanding both tourists’ perceptions and sustainability challenges in saturated destinations. Table 5 Results of the clusters. Class 1 2 3 4 Number of objects by cluster 5 16 4 2 Sum of weights 364 290 225 470 Within-cluster variance 57.031 124.111 57.461 2881.108 Minimum distance to centroid 0.209 0.901 0.736 3.174 Average distance to centroid 0.727 2.293 0.978 3.498 Maximum distance to centroid 3.386 10.021 1.805 3.896 Cluster centroid: User_AI 5.918 6.555 5.646 6.799 Cluster centroid: User_opinion 6.308 6.993 6.027 6.983 Cluster centroid: KD-TAC 21.129 8.766 13.293 32.857 Fig. 5. Silhouette scores. C. Foronda-Robles et al. Annals of Tourism Research Empirical Insights 6 (2025) 100184 8 Consistent with previous studies (Mihalic, 2020; Papadopoulou et al., 2023; Soltani-Nejad et al., 2024), the results confirm that overtourism negatively affects sustainability by increasing economic and environmental pressures. However, this study also shows a positive association between user satisfaction and KDs such as “expensive” and “price”, suggesting that despite cost concerns, the tourist experience in Granada is highly valued. These results indicate tourists’ willingness to pay more for unique experiences (Huang & Chelliah, 2024; Ryu & Kwon, 2021). The concentration of problems in the city centre, especially in commercial areas, confirms the fragility of certain spaces under the pressure of tourism. This spatial imbalance, reflected in KD distribution and cluster groupings, requires management strategies to promote equitable tourist flow distribution and protect vulnerable areas. The results align with those of other studies (Yu & Egger, 2021), emphasising the importance of geospatial tools in identifying dissatisfaction patterns in user comments. The SNA has enabled a visualisation of the interconnections between the tourist attractions and the KDs, identifying critical points that concentrate multiple problem. Albayzin, Capilla Real, Sacromonte and Carrera del Darro, being emblematic attractions and highly centralised in the tourism network, face particular challenges in overtourism management. The findings reinforce the need to implement specific management measures at these critical points, such as flow regulation, diversification of the tourism offer and the promotion of responsible practices [36]. The methodology, combining emotion analysis, multivariate analysis and geospatial tools, offers an approach that is applicable to other tourist destinations. It enables research to move beyond merely describing problems, facilitating an understanding of overtourism and evidencebased decision making. These methodological advances address gaps in the literature, enhancing the understanding of overtourism as a multidimensional phenomenon influenced by economic, environmental and social factors (Buitrago-Esquinas et al., 2023). It is vital to consider tourism congestion as a multidimensional phenomenon influenced by factors beyond the number of visitors. Management capacity, the spatial distribution of tourists and competence at the attractions, as demonstrated by the variability of the KDs and the formation of clusters are fundamental issues that must form part of integrative strategies to deal with overtourism (Yan et al., 2024). Finally, the study opens new research avenues on the perception of overtourism, the elasticity of tourism demand and destination management in the digital age. 6. Conclusions This study shows how tourists’ perceptions, reflected in UGC, can be used to evaluate complex dynamics such as overtourism. This approach aids understanding of the relationship between visitors’ experiences and destination quality perception. The study offers an innovative perspective on how perceptions can help affect destination sustainability, through sentiment analysis and spatial patterns. In addition, analysing overtourism’s impact on the tourist experience expands sustainability theory, providing a solid basis for policies balancing visitor satisfaction and heritage protection. The study also strengthens tourism governance by highlighting the need for regulatory frameworks that balance tourism pressure with destinations’ carrying capacity. Moreover, online reputation is established as a tool to identify specific problems and improve decision making in destination management. 6.1. Practical implications In practical terms, the study has direct implications for tourism operators and those responsible for destination management: Online reviews can be used to identify patterns in tourists’ complaints, which allows managers to adjust services in a proactive way, improving key aspects of tourism such as customer support, cleanliness and the quality of the activities offered. Furthermore, the implementation of regulatory measures such as limiting access to destinations and the use of flow management technologies can help control congestion in areas affected by overtourism and thus improve the tourist experience. Greater control of tourist flows, through mechanisms such as advanced booking, additional fees and even temporary closures, can also help mitigate environmental degradation. The promotion of alternative routes and improvements in urban infrastructures, such as the transport system and basic services, are key interventions that enable the negative effects of overtourism to be mitigated and at the same time the distribution of tourist flows to be improved. Specifically, the improvement of public transport systems and the creation of infrastructures that support a greater number of visitors reduce the economic and environmental impact. Also, the promotion of diversification to encompass other less well known and centralised attractions enables a more equitable tourist flow redistribution. Promoting lesser-known attractions and decentralising tourism to other areas—that is, other neighbourhoods and monuments—is how this more equitable redistribution of tourist flows can be ensured. Such diversification helps prevent overcrowding at major sites while offering tourists new experiences. Tourism operators can also design staff training programmes that are oriented towards the management of large numbers of tourists, making sure that service quality is not affected. Finally, increased collaboration among all stakeholders (local authorities, businesses and residents) can minimise the negative effects of excessive tourism. This collective effort ensures the implementation of sustainable practices and balances the interests of all parties involved. 6.2. Social implications This analysis of overtourism impacts and this use of online reviews to identify recurring problems have significant social implications. The degradation of and congestion at tourist attractions give rise to the creation of natural and cultural conservation initiatives, as well as the promotion of responsible tourism practices. Awareness campaigns based on the insights obtained can redirect tourist behaviour towards a more respectful interaction with the destination, which benefits both local communities and visitors. Additionally, the improvement of urban infrastructures and the promotion of alternative routes not only mitigate the impact of overtourism, but also improve the local residents’ quality of life by reducing pressure on public services and spaces. Appropriate online reputation management also has an important social repercussion, since it enables problems to be dealt with before they negatively affect the destination’s image in the long term. In conclusion, this study provides valuable insights into the impact of overtourism on Granada, using online reviews to identify dissatisfaction patterns related to cost, environmental degradation and social congestion. The combination of multivariate analysis, geospatial tools and sentiment analysis validates and expands existing theoretical frameworks of overtourism, demonstrating how UGC can indicate sustainability challenges. The findings fill critical gaps in the literature, offering a multidimensional perspective on overtourism that integrates economic, environmental and social dimensions. CRediT authorship contribution statement Concepci´ on Foronda-Robles: Writing – review & editing, Writing – original draft, Supervision, Software, Resources, Project administration, Investigation, Funding acquisition, Conceptualization. Luis GalindoP´ erez-de-Azpillaga: Writing – review & editing, Writing – original draft, Validation, Supervision, Software, Methodology, Investigation, Formal analysis, Conceptualization. Pablo Armario-P´ erez: Writing – C. Foronda-Robles et al. Annals of Tourism Research Empirical Insights 6 (2025) 100184 9