Introduction

The Belt and Road Initiative (BRI) marks a decade of significant contributions to global connectivity and urban development since its initiation. According to the white book published by the State Council, by 2023, over 150 countries had joined the BRI, with completing contract projects worth $130 billion in member countries1. The BRI’s scale and impact are profound, particularly in enhancing infrastructure and connectivity in developing regions, which are experiencing rapid urbanization2,3.

BRI projects, encompassing ports, industrial parks, railways, and bridges, integrate deeply into urbanization processes, especially in developing countries4. These projects are crucial for improving local infrastructure and fostering economic growth in regions with fast-growing urban populations5. The integration of BRI projects into urban development highlights their role in reshaping urban landscapes and influencing local communities6.

Despite the broad reach of BRI, there is a notable research gap in quantitative studies that utilize crowd-sourced data to analyze local perceptions of these projects7. As traditional studies have relied heavily on news media and interviews, direct feedback from local communities is limited in small group6,8. Understanding these perceptions comprehensively in a more comprehensive way is crucial, as it offers insights into the real-world impact of BRI projects on urbanization and infrastructure development. However, previous studies suggest that on social media platforms such as Twitter and Facebook, discussions about the BRI are heavily influenced by political dynamics and media narratives rather than the perceptions of actual users or the real-world performance of BRI projects at the local level7. Moreover, Chen et al.56 analyzed a global Twitter dataset on the BRI topic and found that over half of the tweets originated from high-income countries, which do not align with the primary regions of BRI operations, predominantly in developing countries. Given this discrepancy, Google Maps reviews serve as a more targeted and geographical-balanced data source for assessing the local impact of BRI infrastructure projects. Unlike social media discourse, Google Maps reviews are contributed by actual users who have personally visited these locations, providing more direct and experience-based insights into how local communities perceive and are affected by these projects.

Therefore, this study explores the influence of varying categories of BRI projects and regional urbanization rates on the dynamics of public discourse, by analyzing 144,210 Google reviews from 352 BRI urban infrastructure projects between 2012 and 2023. It employs the Valence Aware Dictionary and Sentiment Reasoner (VADER) for sentiment analysis and Multi-grained Latent Dirichlet Allocation (Mg-LDA) for topic modeling. By examining user-generated data, this research provides a spatiotemporal perspective on public sentiment and discussion topics related to BRI projects. This approach aims to offer valuable insights for investors and planners involved in global infrastructure development.

Literature review

Impact of BRI projects on urbanization

Previous studies have investigated the urbanization rates and growth of countries along the Belt and Road. Research conducted by governmental agencies suggests that these countries generally have lower urbanization rates than the global average9, yet many are experiencing rapid urbanization, becoming hotspots of global urban growth10,11. However, a wide urbanization disparity exist among BRI countries. For example, Zheng et al.3 examined three BRI projects of varied spatial scales and geographic backgrounds, arguing that BRI’s effects on urbanization vary. In middle and low-income countries, BRI projects drive local urbanization. In upper-middle-income countries, such as the Belgrade-Budapest railway, they act as catalysts for urbanization. In high-income countries, BRI projects stimulate commerce.

Researchers also explored mechanisms through which BRI influences global and local urbanization through substantial infrastructure investments. Williams et al.12 argued that BRI is creating a unique form of urbanization with a profound global influence, suggesting it should be regarded as a significant urban issue. Other scholars13,14 warned that BRI projects might lead to gentrification, spatial fragmentation, and social segregation, potentially unbalancing urban environments.

Some scholars examined the temporal dimension of BRI projects on urbanization. For example, Ma provided initial evidence that joining the BRI positively affects the urbanization rate of host countries using a difference-in-differences (DID) model, though this impact becomes evident after five years15. Li et al. further proofed that the ports of Maritime silk road enhance the urban expansion of their host cities which is more significant than its effect at country level16.

Moreover, some studies focus on the interaction between BRI projects and local political, cultural, and economic institutions. Through detailed case studies, Goodfellow and Huang examined China-invested transportation infrastructures and industrial parks in Africa, arguing that the performance and outcomes of these projects depend heavily on interactions between private or state-owned entrepreneurs and African governments6,8. This integration into the local urban context results in contingent and unpredictable outcomes, a view supported by studies of China-invested industrial parks in India17 and Malaysia18.

Overall, these studies fall into two categories: quantitative research on urbanization indexes in BRI countries, and case studies on the mechanisms of BRI and urbanization. The former is mainly broad and country-level, while the latter lack quantitative analysis. Beyond urbanization indexes and politics, public sentiment and discourse, as captured through big data, provide valuable insights into cities and infrastructure19,20. Analyzing public perception and its interaction with the local environment is an increasingly popular method for studying urban transformation21,22, but has not yet been applied to BRI urbanization.

Leveraging big data to analyze public perception of BRI projects

Goodchild pointed out that every individual in a city can as act as an agent of urban transformation19. The popularity of social media, location-based social network (LBSN), and big data management has enabled the exploration of public perception from millions of people23. The abundant of data enhances the understanding of urban life, providing practical insights for decision-makers and urban managers21. Machine learning methods such as sentiment analysis and topic modeling are particularly useful in this context24.

By analyzing sentiment and topics over time and space, researchers can understand urban dynamics and the impact of urban attributes on citizens’ lives and sentiments25,26. Applied to urban infrastructure, these methods reveal not only the strengths and weaknesses of facilities but also their interaction with the urban context and overall city impact20,22. For example, research using TripAdvisor to study visitor sentiments about parks and tourist attractions27,28 linked discussion topics to design and planning elements, offering targeted suggestions.

From a temporal perspective, ANOVA analysis has been employed to compare sentiment changes in Google Map reviews of airports29, while Park et al. used fixed effects and long-term panel data to study the association between socio-economic conditions, urban topics, and sentiment30. Overall, analyzing topics and sentiments has become an increasingly popular and important method in urban and infrastructure studies.

While there is limited research on user sentiment regarding BRI infrastructure, several studies have examined media sentiment on BRI across multiple countries by analyzing newspaper reports and articles. Garcia-Herrero and Xu conducted a global sentiment analysis of media coverage in 130 countries, finding that media sentiment on BRI is generally positive except in South Asia31. Trade and investment were the main topics in media articles. Specific studies on Poland, the United Kingdom, and Spain32, as well as sub-Saharan Africa (Nigeria and Ethiopia)33, also indicated predominantly positive media sentiment, although security and debt concerns were prevalent. Additionally, Eastern Europe show neutral media sentiment, with anti-government media showing negative attitudes and pro-government media showing positive attitudes32. This suggests that media perspectives on BRI are highly influenced by geopolitical factors, making it challenging to obtain objective views from media sources alone. Chandra, Cambria, and Nanetti used deep learning to analyze Twitter data mentioning BRI, finding that positive emotions were predominant7. The main topics on Twitter included the economy, environment, and security. These studies highlight the variability and controversy of media sentiment on BRI, influenced by geopolitical factors.

As such, this research investigates 352 BRI urban infrastructure projects to understand how public sentiments evolve and relate to changing characteristics of the built environment. Specifically, we examine three research questions: (1) What topics are being discussed in BRI projects and the associated public sentiments and perceptions? (2) How perception varies in different categories and countries of varied development level? (3) Does the urban context influence public sentiment and topic of interests?

Data and methods

The research consists of three steps (Fig. 1). The first step involves collecting Google Maps reviews of 352 projects (Supplementary materials 1), resulting in a dataset containing 214,475 reviews. Data collection and preparation are detailed in Sect. 3.1. Next, an iterative process identifies clustered topics within each category, yielding 124 topics that are then clustered into four main topics. Additionally, a sentiment analysis tool quantifies the sentiment of the reviews, as described in Sect. 3.2. Finally, ANOVA analysis and fixed effects models are used to analyze the association between urbanization and the results.

Fig. 1
figure 1

Framework of research design.

Full size image

Data collection

The BRI projects can be categorized by their breadth and depth of territorial and embeddedness, financing methods, as well as its cultural sensitivity34. Transformative projects like railways have deep and wide territorial embeddedness and high institutional and cultural sensitivity. The embeddedness of industrial parks varies by scale and industry. This study collects multi-category projects of various scales, forms, investment amounts, and financing methods from three datasets (Table S1), to examine how different types of BRI projects generate sentiment and topics and interact with the local environment.

Specifically, 352 projects are covered, including 37 port projects, 49 industrial parks, 120 railway projects, 37 bridges, 73 stadiums, and 36 mixed-category projects (Mixed), as shown in Table S2 and Fig. 2. These projects were selected according to the process illustrated in Fig. S1. Over 60% of these projects are located in lower-middle-income countries (LM), while only 7% are in high-income countries (H), primarily ports and industrial parks. None of the railway projects are located in high-income countries (Fig. S2a in Supplementary material 3).

Fig. 2
figure 2

© OpenStreetMap contributors, GIS User Community, https://services.arcgisonline.com/ArcGIS/rest/services/World_Topo_Map/MapServer ).

The location of study projects(Software used to develop the map: ArcGIS Pro, https://www.esri.com/en-us/arcgis/products/arcgis-pro/overview; Base map: Sources: Esri, HERE, Garmin, Intermap, INCREMENT P, GEBCO, USGS, FAO, NPS, NRCAN, GeoBase, IGN, Kadaster NL, Ordnance Survey, Esri Japan, METI, Esri China (Hong Kong),

Full size image

Google Map reviews are commonly used in social sensing studies23 and are considered reliable for reflecting users’ sentiments and perceptions35. We used Python to fetch Google Map POI information for these projects, then used the Outscraper website to fetch the reviews. The study strictly adheres to data privacy protection standards by utilizing only publicly available user-generated content from Google Maps reviews. Since these reviews are openly accessible on the platform and do not contain personally identifiable information beyond what users publicly share, the study does not involve the collection of private or sensitive data. We collected 243,800 ratings and 87,990 text reviews (Table S2). Most ratings ranged from 4.0 to 4.5 (Fig. S2b). The reviews were in over 100 languages, with the most common being English (55%), Spanish (11%), French (8%), and Indonesian (4%). Additionally, the association of sentiment and topic with the urbanization context of BRI projects was examined at the county and local levels. The data source of urbanization processes can be found in Table S3.

Topic modeling and interpretation

The review texts for each project category were analyzed using Multi-grain Latent Dirichlet Allocation (Mg-LDA), yielding approximately 20 topics per category, resulting in a total of 124 topics (Supplementary Material 2). By synthesizing the content from the 20 terms with the highest term weights and 200 representative reviews with the highest topic weights for their corresponding topic28, we categorized these 124 topics into four major classes: professional function, service industry, benefit (or disbenefit), and development (Table 1). The classification was respectively conducted by three co-authors and the result was cross validated to ensure the consistency. It allows us to identify common themes discussed by users across different project categories and income levels. The topic modeling process is detailed as follows.

Table 1 Topic description.
Full size table

Because the original review texts were in multiple languages, they were translated into English using Google Translate through Google Cloud Shell. Despite some systematic differences and limitations, machine learning translation is considered an acceptable and effective method in research36. Google Cloud Translation benefits from a vast multilingual training dataset and continuous model updates, ensuring improved accuracy even for low-resource languages. To enhance translation reliability, we implemented a pre-processing step to filter out excessively short or meaningless reviews that lack substantive content. Additionally, after translation, we conducted a manual verification process on a sample of translated reviews to identify potential inaccuracies and mitigate machine translation biases. This approach ensures that the sentiment analysis reflects the original user intent as accurately as possible, reducing noise introduced by automatic translation. Following translation, all reviews were cleaned and converted into appropriate bag-of-word representations using the steps outlined by Arun37. The Python packages gensim and nltk were employed for this process.

This study utilized Mg-LDA for topic modeling. Mg-LDA is an enhanced version of the Latent Dirichlet Allocation (LDA) method, which is commonly used to discover common topics in text38. However, compared to LDA, Mg-LDA is more suitable for analyzing online user reviews and multi-place studies39. In this study, both methods were employed, and their results were compared. The results from Mg-LDA were found to be more accurate and readable for this specific study. The Python package tomotopy was used to conduct Mg-LDA modeling.

For sentiment analysis, the Python package Valence Aware Dictionary and Sentiment Reasoner (VADER) (https://github.com/cjhutto/vaderSentiment) was used to analyze the sentiment of each review. The sentiment index ranges from -1 to 1. An index less than -0.05 indicates a negative review, between -0.05 and 0.05 indicates a neutral review, and greater than 0.05 indicates a positive review.

One-way ANOVA and fixed effect

Because the Belt and Road Initiative (BRI) is an international development project, we adopted the World Bank’s income level classification to indicate the development level40 and categorize infrastructure projects based on their host countries. This classification is widely recognized and frequently used in international development and urban studies41, as well as by government and development aid agencies, making it a more empirically ground and policy relevant framework compared to alternative indices. Host countries were grouped into four categories: high income (H, gross national income [GNI] per capita > $14,005), upper-middle income (UM, GNI between $4516 and $14,005), lower-middle income (LM, GNI between $1146 and $4515), and low income (L, GNI < $1145)40.

A one-way ANOVA was employed to reveal significant differences among these groups in a direct and interpretable manner. In addition, a fixed effects model was applied to analyze the panel data and account for unobserved heterogeneity across countries.

Results

Location situation

Figures S3 and S4a shows that the average rate of urbanization for countries hosting BRI infrastructure from 2016 to 2022 is 46.58%, which is lower than the global average of 54.93%. However, the growth rate of urban population (GRUP) for these countries is 2.93%, significantly higher than the global average of 1.9%. This indicates that most BRI projects are located in rapidly urbanizing countries. Figure S3b illustrates that the majority of BRI projects are situated in urban centers in 2015.

When examining the temporal changes in reviews, Fig. 3 reveals that the earliest reviews of BRI projects date back to 2011, but the number of reviews was very limited before 2016. From 2011 to 2015, there were approximately 500 reviews in total. However, the number of reviews surged to 1,595 in 2016, due to the increasing popularity of Google Map Reviews42 and the implementation of 131 projects between 2016 and 2019, which accounts for over 30% of all projects. With the rise in the number of reviews, the sentiment and topic weights became more stable after 2016. Initially, the topic weight for development was higher but decreased to a more stable proportion as the other three topics increased, particularly for ports and stadiums. This indicates that, before 2014, development topics were more prominent. Despite a drop in the review numbers during the pandemic, there was a rapid increase in reviews post-pandemic, with most projects receiving more reviews (Fig. S5). This trend is similar to the study by Li et al.29 on airports in America. Notably, the reviews of all categories experienced an obvious fluctuation during and after the pandemic (further analysis can be found in Supplementary material 4). For example, railway projects surged significantly post-pandemic, especially for the Lahore Orange Line in Pakistan, which began operations in October 2020.

Fig. 3
figure 3

The quarterly time trend of review and topic.

Full size image

It is also interesting to notice that the sentiment varied in seasons. The ANOVA analysis shows that the sentiment of the third quarter (0.33) is significantly lower than other quarters (p < 0.001). Especially for port and industrial park projects, their third-quarter sentiment (0.35 and 0.27) is respectively 12.5% and 16% lower than their average sentiment (0.40 and 0.32). At the same time, the weight of the topic of professional function increased with decreasing sentiment. The possible reason could be the hot weather, causing the decrease in the patience and endurance of the operation problem. The common complaints include:“ … But prefer ships and let’s go more slowly. In hot conditions 1 and a half hour delay….” and “It is terribly hot and little shade. The waiting room is hot and uncomfortable and we still don’t know what time the ferry leaves.”

Topic and sentiment

Sentiment of reviews

Figure 4a presents the distribution of sentiment scores for all reviews. The average sentiment score is 0.34, which is higher than 0.05, indicating a generally positive sentiment. The majority of reviews are positive, with negative reviews being the least common. Furthermore, Fig. S6a shows that most BRI infrastructure projects have an average sentiment score ranging from 0.2 to 0.55.

Fig. 4
figure 4

The sentiment score of reviews. (a) Kernel density estimation (KED) of overall sentiment; (b) Sentiment scores of all development levels and categories.

Full size image

A one-way ANOVA revealed that reviews from upper-middle-income countries (UM) have the highest average sentiment score of 0.42 (p < 0.01), while reviews from high-income countries (H) have the lowest average sentiment score of 0.35 (Fig. 4, b). Additionally, reviews from high-income and upper-middle-income countries have a higher proportion of negative sentiment and greater standard deviation compared to their counterparts. This suggests that projects in higher-income countries tend to be more controversial.

Sentiment scores vary significantly among project categories (p < 0.01), except for ports and bridges. Mixed projects have the highest average sentiment score at 0.48, followed by ports (0.41), bridges (0.40), stadiums (0.38), and industrial parks (0.32). Railway projects have the lowest average sentiment score at 0.26, with the highest percentage of negative reviews (19.4%), particularly for railway projects in low-income countries.

Topic weight and sentiment

As mentioned in the methodology, the sub-topics can be categorized into four major topics: professional function (44%), benefit (or disbenefit) (24%), service industry (19%), and development (13%). Figures S6b and S7 shows a distinct difference in topic proportions between higher-income and lower-income countries. Users in higher-income countries discuss service industry topics more, while users in lower-income countries focus more on development topics.

To calculate the sentiment related to each topic, we labeled reviews according to the topic with the highest proportion in each review and calculated their sentiment scores. The results indicate that professional function topics have the lowest average sentiment of 0.26, while service industry topics have the highest average sentiment of 0.47 (p < 0.01), as shown in Fig. S8. However, both have high deviations and the negative reviews percentage, indicating they are more controversial. In contrast, development topics have the lowest percentage of negative reviews and deviation but the highest percentage of neutral reviews. In the following sections, we delve into the details of each topic.

Professional function topics

High-income countries have the highest topic weight but the average sentiment is the lowest (p < 0.01). Additionally, users of railway projects focus on this topic the most, and its sentiment decreases with the income level. For most BRI projects, the professional function topic accounts for 40–60% of the discussion. Among railway projects with a high proportion of this topic, the Jakarta-Bandung High-Speed Rail Project in Indonesia stands out as an example (Fig. 5). As a flagship project connecting China’s Belt and Road Initiative with Indonesia’s "global maritime fulcrum" vision, this project is the first high-speed railway in Indonesia and Southeast Asia, having come into operation in October 202343. One typical review states: “The station is new, large, clean, modern, although it is still not filled with public service tenants…” However, some railway projects in lower-middle-income countries, such as certain stations on the China-Laos railway line, exhibit low sentiment on this topic. This disparity suggests that the professional performance of BRI railway projects is highly influenced by the local environment of the host countries.

Fig. 5
figure 5

The proportion and weight of topics at project level.

Full size image
Service industry topics

We find that high and upper-middle-income countries tend to discuss service industry topics more frequently (Fig. S6). In contrast, lower-middle-income countries discuss these topics the least and exhibit the lowest related sentiment. Port projects, such as Walvis Bay Port in Namibia and Darwin Port in Australia, are especially prominent in discussions of service industry topics. Additionally, some industrial parks also feature active service industries, such as the Morowali Industrial Park (IMIP) in Indonesia. IMIP, covering an area of 3200 hectares, focuses on nickel metal mining and smelting, and stainless-steel smelting44. This industrial park has evolved into a complex industrial town with ports, airports, restaurants, and hotels. The prominence of the service industry topic suggests that BRI projects not only fulfill their primary functions but also serve as hubs for shopping, travel, visiting, and entertainment for local communities.

Benefit topics

Benefit topic proportions are similar across income levels, but high-income countries have the lowest sentiment (p < 0.01) and highest negative review percentage (9%), which decreases to 4% in low-income countries. Additionally, stadium and bridge projects have a high benefit topics proportion because the advantages of these projects are easily and directly felt by the users.

For example, bridges can improve local traffic, reduce commute times, and offer beautiful and scenic views. The 6th Bangla-China friendship bridge in Bangladesh illustrates this well with reviews such as: "Before evening to midnight, many people come to visit this bridge to enjoy the refreshing air of the river, enjoying the sunset view, etcetera…" and "It has set a unique precedent for the fastest time from Sadar Munshiganj to Narayanganj, Dhaka …."

As for stadiums, they not only provide residents with a venue to watch big events and concerts but also serve as places for leisure activities such as physical exercise and family play. Moreover, the spectacular outlook of the stadiums attracts locals and changes the urban landscape. There are comments such as: "A wonderful and relaxing place for families and a place for children to play." Furthermore, although projects with a high proportion of benefit topics are mainly in middle-income countries, some are in high-income countries, particularly Small Island Developing States such as the Maldives (China-Maldives Friendship Bridge) and Barbados (Cheapside Market). These countries have limited infrastructure construction capacity and face unique social, economic, and environmental vulnerabilities. It is understandable that they are particularly sensitive to the benefits of the BRI projects.

Development topics

Lower-income countries discuss development topics more frequently. Interestingly, lower-middle-income countries have higher sentiment scores on this topic compared to low-income countries. Many reviews in low-income countries are neutral factual statements, such as “Multinational container transport company from the autonomous port of Lomé to serve the hinterland.” In contrast, reviews from lower-middle-income countries are more emotional and abundant. Most projects with a high proportion of development topics are also in lower-middle-income countries, often as flagship projects of the Belt and Road Initiative, such as Gwadar Port in Pakistan, Padma Bridge in Bangladesh, and Lekki Free Trade Zone in Nigeria.

For example, Gwadar Port is frequently mentioned for its regional importance within the Belt and Road Initiative and China-Pakistan Economic Corridor. Examples include “One of the largest deep-seaports in the world. Major port in the South Asia region. Gwadar is one of the gifts of Almighty Allah. May Almighty Allah always protect Pakistan.”

Overall, bridges, stadiums, and industrial parks have a relatively higher proportion of development topics. Padma Bridge, which opened in September 2023 in the capital of Bangladesh, is called “the bridge of dreams.” It is predicted to fundamentally change local transportation and the economy45. Reviews express excitement and hope, with comments like, “The dream of the people of the southern region. Symbol of Bengali pride. ♥♥♥”. Reviewers also see stadiums as symbols of modernization and internationalization, representing national development and progress. This aligns with Dubinsky46, who found that Chinese-assisted stadiums "promote a sense of national progress for citizens and the world", reflecting the desire of recipient countries to host major events.

Association between perception and urban context

Table S4 highlights that the topic proportion and sentiment of BRI projects are significantly influenced by the urbanization conditions of host countries, especially the GRUP. Regarding the topic proportion, the GRUP significantly increases development topics in industrial parks while decreasing benefit topics to a similar degree. This suggests that visitors to industrial parks in rapidly urbanizing countries are more concerned with development issues than with immediate benefits. For instance, the Lekki Free Trade Zone in Nigeria illustrates this trend, with a development topic proportion of 0.34 compared to a benefit topic proportion of 0.14, approximately three times higher. In contrast, examples like the Hualing Industrial Park in Georgia show a higher benefit topic proportion (0.24) and a lower development topic proportion (0.06). Conversely, stadiums in countries undergoing rapid urbanization tend to focus more on benefit topics. Furthermore, Table S5 which replaces country urban population rate into country development index demonstrates a strong association between the sentiment and the proportion of development topics for railway and industrial park projects with the GRUP, underscoring the critical impact of these projects in rapidly urbanizing countries.

Moreover, we can find that both the urban population rate and the GRUP positively impact the sentiment of BRI projects, particularly for those during periods of rapid urbanization. Specifically, GRUP has a notable positive effect on the percentage of positive reviews. While the urban population rate can increase both positive and negative reviews, its effect on positive reviews is twice as pronounced as on negative ones. This finding aligns with previous results indicating that sentiments in higher-income countries tend to be more polarized. Therefore, users in countries experiencing rapid urbanization and those with established urban bases show higher satisfaction levels with BRI projects. For example, industrial parks and port located in countries with higher urban population rates tend to exhibit higher sentiment and a greater proportion of positive reviews. While stadiums in countries with higher urban population rates generally show lower sentiment levels, the GRUP significantly enhances their sentiment, with an effect parameter of 8.83. However, the sentiment of railway projects, having no significant association with country urbanization, shows a positive association with the urbanization rate within a 35 km buffer zone (Fig. S4b,c). This suggests a link between the performance of BRI railway projects and the support provided by surrounding urban facilities.

Furthermore, to explore the role of investment amount in project performance, we introduced the logarithm of investment amount as a factor into the fixed-effect model in Table S6. The results indicate that higher investment amounts are significantly associated with a higher percentage of both positive and negative comments, with parameters around 0.015. This suggests that projects with higher investments tend to provoke more divergent reviews. For instance, railway projects with substantial investments tend to garner a higher percentage of positive reviews and overall higher sentiment. In contrast, stadiums with high investments often receive more negative reviews and tend to exhibit lower overall sentiment. A possible explanation is that stadiums with higher investments attract more users but also entail higher maintenance costs, which can burden host countries47. The Estadio Nacional de Costa Rica exemplifies this paradox, with an investment of approximately US $100,500 but a 15% proportion of negative reviews. Most of them are due to maintenance issues. For example, some reviews complain that: “…the deterioration due to lack of maintenance is noticeable.” and "What a shame that they broke the doors of the bathrooms, both the men’s and the women’s… What is incredible is that they no longer have a lock or handle and doors are on the floor or some doors have been removed."

Discussion

BRI urban infrastructure and urban context

Some previous studies have criticized BRI urban infrastructure as a threat to urban formation, arguing that it exacerbates gentrification and fragmentation13,48,49,55. However, this study offers a new perspective by examining the role of BRI projects through firsthand user experiences. Our findings indicate that BRI projects serve multiple urban functions and generally contribute positively to host cities, as evidenced by a sentiment score of 0.34 and diverse topics. This contrasts with the sentiment score of 0.11 derived from a Twitter database compiled by Hong Kong University, which also employed VADER to assess sentiments regarding BRI. Notably, over half of the tweets in that database originate from high-income countries. Prior literature has emphasized that discussions on media platforms tend to focus on abstract themes such as investment, trade31, and broader economic, environmental, and security concerns7,50. In contrast, reviews on Google Maps primarily highlight concerns related to daily life. This disparity suggests potentially different perceptions of BRI projects between virtual discourse and real-world user experiences.

Furthermore, our study reinforces the notion that the performance of BRI projects is highly contingent upon the local urban context8,17,28. For instance, industrial parks in rapidly urbanizing countries tend to receive more positive sentiment and a higher proportion of development-related discussions. This aligns with Goodfellow et al.6, who argue that Chinese industrial parks function as key drivers of both urbanization and industrialization in Africa. Moreover, the effectiveness of these parks is closely tied to the specific socio-economic and political conditions of their host countries51,52. Similarly, a study on railway and highway projects in Addis Ababa (Ethiopia) suggests that the urban political environment plays a decisive role in shaping the final outcomes of BRI projects8. This finding explains why sentiment and discussion topics surrounding railway projects are significantly influenced by city-level urbanization dynamics. In this sense, BRI urban projects should be understood as both products of Chinese investment and integral components of local urbanization processes.

BRI urban infrastructure in developing countries

BRI projects engage with cities at various stages of urban development, ranging from dense urban cores in developed nations to emerging urban centers and rural regions in developing countries. This study, leveraging a quantitative analysis of crowdsourced data, confirms that user perceptions of BRI projects vary significantly with the urbanization rate and development level of host cities. For example, developing countries are more interested in development topics, while high-income countries focus on the service industry. These findings can be explained by Zheng et al.3, who argue that BRI projects in middle- and low-income countries serve as catalysts for urbanization.

Notably, the development-related topics remain an important characteristic of BRI projects compared to other infrastructure. An intriguing finding is that the urban population rate of a country is not significantly associated with development topics, contrary to expectations based on HDI. While the average HDI for countries involved in BRI projects is 0.64—lower than the global average of 0.73—many of these nations experience "urbanization in poorer countries" where urbanization rates exceed overall development levels53. For example, Ethiopia and the Democratic Republic of Congo exhibit high urbanization rates despite facing severe infrastructure deficits and widespread slum conditions53. This context likely explains why users express heightened concern about the developmental aspects of BRI projects.

Additionally, while the proportion of development-related topics is similar between middle- and lower-income countries, reviews from lower-middle-income countries tend to be more emotional and exhibit a more positive sentiment. Additionally, lower-income countries host the majority of BRI projects. Ma’s study also underscores that BRI significantly impacts the urbanization trajectories of lower and middle-income countries15. These findings underscore the critical role of BRI projects in developing and transforming countries.

Policy implication

The findings underscore the importance of aligning aid projects with the contextual realities of host countries, a critical factor influencing project outcomes54. These results suggest that investors and planners of BRI projects should carefully tailor designs and plans according to the development level and genuine needs of the host countries.

In lower-incomes countries, caution should be exercised regarding high-investment projects requiring intensive maintenance. Both investors and recipients need to consider long-term maintenance costs carefully. The study supports the “small yet beautiful” turn in BRI 2.03, advocating practical and durable design to minimize long-term maintenance challenges.

Moreover, the study highlights that user sentiment is significantly higher in areas with higher urbanization rates, particularly for railway projects. Therefore, urbanization rates and growth dynamics should be considered in site selection to enhance integration with surrounding communities.

For projects such as ports and industrial parks, which exhibit a lower proportion of benefit-related topics indicating fewer tangible benefits to users and isolation from the local community, the study suggests focusing on environmental stewardship and creating aesthetically pleasing and accessible environments. Designers should prioritize protecting local environments, and planning open and attractive spaces that help for integrating with the local community. For example, considering the prevalent reviews about the hot environment and lack of shelters in the summer, this study that ports and industrial parks enhance greenery and provide additional shelters to improve thermal comfort.

Limitations and further research

One limitation of this study is the challenge in defining the impact scope of some projects and capturing indirect influences. BRI urban infrastructure projects can have extensive external effects on entire host cities and regions, but this study focuses solely on direct reviews of these projects. Future research should aim to comprehensively assess the broader impacts of these projects on host cities and regions.

Another limitation is the reliance on data from Google Maps reviews. Although Google Maps is one of the most popular maps globally, the penetration of Google services varies across regions, with lower adoption rates in certain low-income countries. While this study represents a pioneering effort in using crowd-sourced and social sensing data to analyze BRI projects, it may overrepresent users of smart devices and those using Google products, particularly in lower-income countries and rural area. As a result, this study has limitations on capturing the sentiments of less active or non-digital participants which is prevalent in many BRI project locations. Future research could mitigate this limitation by integrating crowd-sourced data with traditional investigative methods, such as interviews, providing a more balanced and nuanced understanding of the impacts of BRI projects.

Furthermore, this study focuses on the impact of urbanization and income levels on the perception of BRI projects and found their significant effect but does not involve other potentially significant factors such as built environment, urban political environment and social-cultural context. There remains substantial room for incorporating these dimensions through more refined analytical models. Further studies are suggested to explore the mechanisms behind the public BRI infrastructure perception through interdisciplinary collaboration across various fields and domains.

Conclusion

Our study moves beyond the conventional approach of analyzing the public perception of the BRI through media or Twitter data. Instead, we innovatively utilize Google Maps reviews to examine users’ firsthand experiences with BRI infrastructure and how this infrastructure functions as a place of users’ daily lives. Furthermore, by engaging with the ongoing debate on "BRI urbanization," we explore how different levels of urbanization, in conjunction with BRI infrastructure, shape user and community perceptions and everyday experiences. Our findings indicate that user reviews predominantly reflect positive sentiment, averaging 0.34 (ranging from -1 to 1). Besides, discussions among users primarily revolve around four key topics: professional function, service industry, benefit (or disbenefit), and development. Different from the previous studies using Twitter and media data, professional function about daily operations emerges as the most prevalent topic whereas development is discussed less frequently. Furthermore, infrastructures situated in countries with higher urbanization rates and urbanization growth rates tend to receive higher sentiment scores. However, projects with substantial investments may experience more polarized sentiments.

Based on the novel findings, it provides actionable insights for investors, planners, and local primary governments. Emphasizing the importance of integrating urbanization dynamics into strategic planning processes, the study underscores the need for tailored approaches that address the specific challenges and opportunities posed by BRI infrastructure projects in developing and transitional contexts.