Abstract
Introduction:
Urban mobility has emerged as a key factor in the challenge of achieving social and environmental sustainability in major Latin American metropolises such as São Paulo (Brazil), whose metro system has a high passenger density and an infrastructure that does not always meet user demands. In contrast, the Metro do Porto (Portugal) seeks to maintain high satisfaction rates in the face of challenges posed by growing tourism in the city and pressure on its system.
Methods:
Therefore, the methodology of this study conducts a longitudinal comparative analysis between these contexts, introducing the concepts of “Public Value” and “Travel Experience,” focusing on the evaluation of the passenger experience and sentiment analysis based on User-Generated Content (UGC) data collected from Google Maps between January 2017 and December 2025 through an automated extraction system. The analysis of the comments was conducted using Natural Language Processing (NLP) and Sentiment Analysis, employing DeepSeek-V3, a Large Language Model (AI model), to correlate users’ perceptions of the system with travel experience metrics.
Results and discussion:
The results show a positive perception of both systems. It was noted that, despite the pressure caused by the growth of tourism and the problems identified with the ticket machines, the Trindade station continues to receive positive reviews, while the Sé station stands out for its efficiency and intermodal integration, even amid security crises. This analysis enables the development of safer and more inclusive metro systems, in accordance with the United Nations (UN) Sustainable Development Goals (SDGs) 9.1 and 11.2.
1 Introduction
This study analyses user satisfaction with two metro stations in the cities of Porto, Portugal, and São Paulo, Brazil, based on the sentiment expressed by users in Google Maps reviews and processed through an artificial intelligence model (DeepSeek-V3). Addressing their operational characteristics, statistical data, historical context, and the challenges inherent to their infrastructure and expansion. The comparison aims to highlight the particularities of each system, considering their respective urban environments and mobility demands.
discusses post-pandemic urban life and how urban mobility has become central as a tool for citizenship. Because public transportation enables economic activities and promotes integration among the population and regions, while facilitating access to essential services, employment, and education, it significantly contributes to sustainable urban development.
Metro transportation in major metropolises, such as São Paulo, represents an critical transportation system whose functional performance directly impacts the population’s quality of life, operational efficiency, and urban sustainability (; ; ). Porto is the second largest city of Portugal and its metro system is recognized for its efficiency and sustainability; in its most recent survey, it achieved an excellent customer satisfaction score (85.7/100 points) according to . Thus, studying a system with high satisfaction levels allows for the identification of best practices in management and operations ().
This paper aims to survey and analyze the sentiments of users at Sé Station, located in São Paulo (Brazil), and Trindade Station, located in Porto (Portugal), regarding metro service quality, as well as their perception of the travel experience and public value in transportation. Using User-Generated Content (UGC) from Google Maps and Sentiment Analysis using DeepSeek-V3 Artificial Intelligence (AI) model, the study seeks to understand user trends and needs. This approach identifies strengths and critical areas for improving the system and the travel experience based on user perception, ultimately contributing to the development of safer transportation that provides a higher quality of travel. From a theoretical perspective, the study empirically operationalizes the concepts of Public Value and Travel Experience through UGC, integrating them into an analytical model applied to two distinct socioeconomic contexts.
2 Theoretical foundation
2.1 Metro public transportation as a social and environmental driver
The city of São Paulo depends directly on efficient public transportation—not only focused on reducing traffic and travel time but also serving as a tool for urban social inclusion, where users from diverse social classes, through regional integration, have access to employment, educational, and healthcare opportunities outside their usual region ().
However, public transportation also stands as an alternative for reducing pollutant emissions compared to private vehicles, highlighting its relevance in the face of climate crises. Despite challenges related to user demands, the metro system serves as a barometer of urban development in functional cities (). According to the and , the world population, starting in 2024, will grow by 1.5 billion by 2050, and the urbanization rate will reach 67%, intensifying the need for efficient urban management.
Given this scenario, the promotion of sustainable urban mobility infrastructure is a global priority that aligns with Sustainable Development Goals 9 and 11 (SDGs) of the United Nations (UN) for 2030, which aim to transform transportation systems into catalysts for social inclusion, energy efficiency, and the reduction of environmental impacts ().
The Metro do Porto began commercial operations on 7 December 2002. Classified as a rapid transit system, its network consists of six lines (A-Blue, B-Red, C-Green, D-Yellow, E-Violet, and F-Orange), totaling 70 km in length and 85 stations, 15 of which are underground. In 2024, the system recorded 89.8 million annual passengers (). Figure 1 shows the network map.
FIGURE 1
The Trindade station connects most of the lines, serving as the system’s main integration hub, with over 17 million passengers recorded in 2024. The Metro do Porto faces several challenges, such as network expansion (including projects for the G-Pink Line and the extension of the D-Yellow Line) rolling stock maintenance, and adapting to growing mobility demands resulting from tourism in the region ().
The São Paulo Metro system in Brazil is older, having begun operations on 14 September 1974. Its metro network consists of seven lines (1-Blue, 2-Green, 3-Red, 4-Yellow, 5-Lilac, 15-Silver, and the recently inaugurated 17-Gold), of which Lines 4 and 15 are operated by private organizations, and Lines 15 and 17 are a monorail system. In total, the network has 91 stations and a total length of 104.4 km (). Figure 2 illustrates the São Paulo station network.
FIGURE 2
The Sé station connects two subway lines (Lines 1-Blue and 3-Red), and in 2025, considering entries, exits, and transfers between these lines, an average of approximately 405,000 passengers were recorded on weekdays, accounting for 22% of the system’s total ridership ().
There is a significant difference in age between the two analyzed systems. While the São Paulo Metro has been in operation for about 51 years, the Metro do Porto is a much younger system, having been in operation for approximately 24 years, which is reflected in their respective infrastructure and operational challenges. The São Paulo system is characterized by its inception during a period of rapid urbanization, operating under pressure to meet a population demand greater than its capacity and without an integrated transport structure (; ).
Although they operate on different scales and in different contexts, both systems face a constant need for improvement and adaptation in the face of population and urban growth in their respective cities. They share common objectives such as expansion into underserved areas, the pursuit of greater modal integration, and sustainability (; ; ; ).
Trindade Station is part of the Metro do Porto; the system is primarily classified as light rail or Light Rail Transit (LRT), operating on both at-grade tracks and underground sections, and is considered the main interchange hub of the network (). While Sé Station is part of the São Paulo Metro, this section of the system is a heavy metro (heavy metro/rapid transit), characterized by high transport capacity and high demand. The station is underground and serves as a major hub between Lines 1-Blue and 3-Red ().
2.2 Population and urban characteristics of the cities analyzed
The Metro do Porto currently serves seven municipalities in the Porto Metropolitan Area: Porto, Matosinhos, Maia, Póvoa do Varzim, Vila do Conde, Vila Nova de Gaia and Gondomar (). In 2024, PORDATA, the database of Portugal, reported the population of these municipalities, which totaled approximately 1,215,099 inhabitants, while the census conducted by the documented that the population of the municipality of São Paulo reached a milestone of 11,451,999 inhabitants, and this population size is one of the main factors creating challenges in the management of its metro system.
Despite the notable population difference between the two cities, a comparison of their metro systems is warranted due to common factors in the context of sustainable urban mobility. São Paulo, with its vast population, represents a model of a global metropolis facing transportation challenges. On the other hand, Porto, with its smaller population, offers an example of how a system can be successfully implemented and operated in a medium-sized city ().
According to , Porto has seen a 36.9% increase in the number of tourists in recent years, averaging 17,204 people per day, which poses constant challenges and requires constant innovation in its transportation systems. This city ranks among the four districts with the highest concentration of foreign residents, with Brazilians constituting the largest foreign community residing in Portugal (; ).
Exposure to a multicultural environment is fundamental for transferring knowledge to the context of Brazilian urban mobility (). This analysis in distinct contexts (Latin America and Europe) does not seek to establish a hierarchy, but rather to identify aspects of the Travel Experience of users in the cities of Porto and São Paulo and how Public Value is established, in order to inspire improvements and solutions adapted to the Latin American scenario, even with its nuances.
2.3 Contextualisation of public value and travel experience
While classic customer satisfaction theory () typically measures the gap between expectations and perceived performance, this study operationalizes user satisfaction through two complementary constructs—Public Value and Travel Experience—which capture not only service performance but also the collective, civic dimension of public transportation, more suited to the comparative, cross-country nature of the analysis.
Beyond operational efficiency and cost-benefit analysis, state that although the concept of Public Value has various definitions, it is also considered the most appropriate tool for validating policies that advance the public interest, since it is treated as the expression of collective preferences, shaped by politicized processes, with economic development, cultural, and social cohesion serving as pathways to create public value. In this context, the perspective of value co-creation (; ) broadens the analytical lens by recognizing that Public Value is not produced solely by the State but is linked to the interaction between managers, operators, and citizens. In the context of public transportation, the subject of this study, users actively participate in this process by sharing their experiences on digital platforms such as Google Maps, which makes User-Generated Content (UGC) not only an evaluation metric but also a tool for diagnosis and legitimacy, constituting a form of citizen participation in transportation governance.
As proposed by , Public Value is not limited to service delivery but lies in the manager’s ability to balance three essential dimensions: the judgment of what is socially relevant and fair (such as accessibility and wellbeing), legitimacy and political support (authorization to act), and operational feasibility (the capacity to deliver results). In this sense, it is important to clarify that, for this study, Public Value focuses on the social relevance and legitimacy of transportation policies, directly linked to infrastructure planning that seeks to mitigate problems related to wellbeing and accessibility (; ; ).
proposes four basic dimensions of value, which are interrelated and derived from fundamental human needs: (1) utilitarian-instrumental, related to the effectiveness and practical outcomes of services; (2) hedonic-aesthetic, associated with quality of life, pleasure, and positive emotional experiences; (3) social-political, focused on social relations, group identity, and a sense of belonging; and (4) moral-ethical, focused on dignity, justice, and respect for people. Of these four layers, this study explicitly addresses three: the utilitarian-instrumental dimension, operationalized through criteria of efficiency, cost-benefit, and accessibility; the hedonic-aesthetic dimension, captured by the physical and sensory experience subdimension of the Travel Experience (comfort, cleanliness, and atmospheric perception of the space); and the socio-political dimension, reflected in the safety and social cohesion indicators present in user evaluations. The Travel Experience, on the other hand, reflects perceptions and behavior during the journey, being decisive for the attractiveness of transportation—linked not only to efficiency but also to sensory aspects of the route taken ().
The moral-ethical dimension, relating to the treatment of individuals with dignity and equity, was not directly operationalized as an isolated analytical category, as the study’s focus is on service performance and experiential quality, rather than on normative assessments of governance. This delimitation does not reduce the analytical scope of the research; on the contrary, it ensures that the dimensions addressed are directly observable through User-Generated Content (UGC) and classifiable by the DeepSeek-V3 sentiment analysis model.
2.4 Sentiment analysis performed by artificial intelligence and user-generated content (UGC)
AI-driven sentiment analysis allows for the interpretation of emotions and opinions contained in public texts, such as online reviews, and the extraction of perception patterns, recurring topics, and temporal trends in satisfaction/dissatisfaction (). In this study, sentiment polarity (positive, negative or neutral) is used as an operational proxy for user satisfaction: positive sentiment indicates a favorable evaluation of the service (satisfaction), negative sentiment indicates an unfavorable one (dissatisfaction), and neutral sentiment indicates the absence of an evaluative judgment. In the transportation sector, this tool has been used to go beyond traditional surveys, capturing in a richer and more qualitative way how passengers actually experience a trip (; Oliveira et al., 2025
Therefore, it is also necessary to emphasize the importance of using User-Generated Content (UGC) for this research, since UGC will serve as the foundation for exploring user interactions and opinions, providing a rich and contextualized view of metro services with the assistance of DeepSeek-V3. UGC is formed through the voluntary participation of a community that shares information and experiences via social media, making their perceptions available; in our case, Google Maps reviews allow us to understand passengers’ needs through their opinions (; ).
The choice of DeepSeek-V3 as a sentiment classifier is justified by two methodological criteria: (i) growing recognition of its global adoption due to its competitive performance in reasoning tasks and multilingual semantic understanding capabilities, as per an independent evaluation by the National Institute of Standards and Technology (); and (ii) its superiority over traditional dictionary-based sentiment analysis models (such as the Valence Aware Dictionary and sEntiment Reasoner—VADER) in texts with contextual ambiguity and regional expressions, as highlighted by . In addition to this, the careful generation of prompts (analysis commands) and validation via Cohen’s Kappa constitute the control protocol adopted to mitigate classification variations arising from idiomatic and regional expressions in sentiment analysis using artificial intelligence and natural language processing (NLP).
3 Methodology
The research focused on reviews extracted from Google Maps for two stations, Sé and Trindade, and their metro systems, located in the cities of São Paulo (Brazil) and Porto (Portugal). For this analysis, the choice of São Paulo was based on factors such as population size and economic importance, while Porto was chosen for the efficiency of its metro services and the expansion challenges posed by the increasing number of tourists and system users. In this study, databases on passenger volume per station, as well as government and institutional reports and indicators, were examined to determine which station in each city had the highest passenger traffic over the course of a year.
To collect the reviews, a program developed by was used, called the “Automated Google Maps Review Collection System,” which enables the automatic and continuous collection of reviews, allowing for the creation of a more comprehensive database. Reviews were extracted for the period from January 2017 to December 2025. The collection protocol restricted the unit of analysis to each individual review, considering the number of stars given by users, the user’s name (ID), whether photos were included, whether the user is classified as a “Local Guide” (participatory users), and the publication date.
A field visit was conducted to Trindade Station, which allowed us to document the use of ticket machines, and images recorded by users (Google Maps) were used to illustrate passenger traffic at Sé and Trindade stations.
It is important to note that, unlike traditional NLP pipelines that require explicit tokenization, stemming, or stopword removal, this study employs a prompt-based classification approach using a large language model (DeepSeek-V3), which performs semantic interpretation directly on raw text.
Sentiment Analysis (SA) was performed using DeepSeek-V3 Artificial Intelligence (AI) (updated in May 2025), based on nine pre-established criteria aligned with the analysis of Public Value and Travel Experience, such as: efficiency, accessibility, customer service, amenities, comfort, cleanliness, infrastructure, fare satisfaction, and safety, requiring attention to factors such as modal integration, cost-effectiveness, speed, and maintenance. Based on this, the following prompt (execution command in Brazilian Portuguese) was developed by the authors and applied to the spreadsheet containing user comments:
You are a classifier specializing in sentiment analysis of public transportation reviews on the Google Maps platform. Classify the OVERALL SENTIMENT of the comment about public transportation based on NINE criteria: efficiency, accessibility, customer service, amenities, comfort, cleanliness, infrastructure, fare satisfaction, and safety. Also consider modal integration, cost-effectiveness, speed, and maintenance.
COMMENT TO BE CLASSIFIED INDIVIDUALLY: [spreadsheet with sample comments (for Kappa calculation)/spreadsheet with all comments from each year for AS].
Rating Criteria Rules:
Overall Sentiment: Answer ONLY with one of the following words: positive, negative, or neutral for:
“Positive”: most applicable criteria are favorable (when the user expresses satisfaction, praise, relief, or a pleasant experience);
“Negative”: most applicable criteria are unfavorable (when the user expresses dissatisfaction, complaint, frustration, or an unpleasant experience);
“Neutral”: there is no clear dominance or only factual comments (without clear emotional tone (e.g., “the subway runs from 6 a.m. to 11 p.m.”)).
Public Value (dimensional analysis): Does the comment demonstrate that the service solves a public problem? Does it generate wellbeing for the community? Classify EACH of the three sub-dimensions as “present” or “absent” (“positive” or “negative” in the case of sentiment analysis):
- ○
Accessibility - present if the comment mentions: “elevator,” “ramp,” “PWD,” “signage,” “inclusion,” “map,” “escalator,” “staircase,” “adapted platform”;
- ○
Safety - present if the comment mentions: “lighting,” “police presence,” “safety/danger,” “security guard,” “fear,” “trust,” “assault,” “theft” (indicating a lack of safety);
- ○
Cost-benefit - present if the comment mentions: “fare price,” “ticket price,” “free,” “integration/connection” between modes of transport, “expensive,” “cheap,” “fair,” “abuse”;
- ○
Public Value of Travel Experience: Physical and Sensory Experience (Body as a sensor) - present if mentioning: “clean/dirty,” “hygiene,” “trash,” “chaotic/calm,” “organized/messy,” “comfortable/uncomfortable,” “cramped,” “stink,” “smell,” “odor,” “stuffy/fresh,” “heat,” “cold,” “ventilation”;
- ○
Service Quality (Designed Space) - present if mentioning: “crowded,” “overcrowded,” “empty/full,” “service,” “staff,” “employee,” “delays,” “punctuality,” “schedule,” “maintenance,” “broken”;
- ○
Emotional and Affective Experience (Lived Space) - present if mentioning: “memory,” “nostalgia,” “recollection,” “longing,” “peaceful,” “stressful,” “calm,” “nervous,” “touristy,” “landscape,” “historic,” “heritage,” “tradition,” “old,” “community,” “landmark.”
- ○
For Public Value, generate rows for the results: one for accessibility, one for safety, cost-benefit, one for Bodily and Sensory Experience (body as a sensor), one for Service Quality (designed space), and another for one for Emotional and Affective Experience (lived space).
Based on the results, generate a report with content and sentiment analysis of the testimonials found. Also analyze Public Value and the Travel Experience, which will be assessed through the user’s direct perception of the environment and service. Analyze the longitudinal report (2017–2025) and perform the following cross-tabulations for the Sé and Trindade stations: Classify each comment with two simultaneous labels: Public Value (PV) (positive and negative) and Travel Experience (TE) (positive and negative). Generate the co-occurrence matrix and display the dissociation percentages. Filter the 1- and 2-star ratings and identify which sub-dimension of TE appears as negative in that comment.
The analysis was performed using the same prompt for each year of each station individually, and at the end, the results were compared across stations.
The operationalization of the analytical dimensions was structured according to Table 1, which represents the study’s conceptual framework, used as the basis for the classification prompt executed in DeepSeek-V3.
TABLE 1
| Analytical layer | Concept/theory | Operationalized dimensions | Indicator via CGU (Google Maps) |
|---|---|---|---|
| Sentiment | Overall | Positive, Negative and Neutral | “Efficiency”, “accessibility”, “customer service”, “amenities”, “comfort”, “cleanliness”, “infrastructure”, “fare satisfaction”, “safety”, “modal integration”, “cost-effectiveness”, “speed” and “maintenance” |
| Public value | Social judgment of the common good | Accessibility | “Elevator,” “signage,” “dignity,” “inclusion” |
| Perceived space | “Clean/dirty,” “bad smell,” “organized/messy,” “comfortable/uncomfortable” | ||
| Conceived space | “Staff,” “delays,” “service,” “crowding” | ||
| Lived space | “Memory,” “landmark,” “calm/stressful,” “touristy,” “historic”/“heritage” | ||
| Political legitimacy | Security | “Policing,” “safety/danger,” “government” | |
| Operational feasibility | Cost-effectiveness | “Fare price,” “integration,” “expensive/cheap” | |
| Method: Longitudinal sentiment analysis via AI (DeepSeek-V3) google maps CGU (2017–2025) | Output: Positive/Negative/Neutral classification by dimension and sub-dimension | Cohen’s Kappa |F1-Score | ||
Framework of analytical dimensions and key terms for sentiment classification and analysis.
Source: Prepared by the authors based on , , , and .
As shown in Table 1, we sought to operationalize the Public Value AS, focusing on three main dimensions extracted from the comments: Accessibility (social judgment of what is fair), Safety (legitimacy/political support), and Cost-Benefit (operational feasibility), in order to identify words indicating that the service solves a public problem or generates wellbeing for the community (; ; ). Additionally, as studied by , the Travel Experience was operationalized through the user’s direct perception of the environment, the service, and the human dimension of travel. This structure allows the DeepSeek-V3 AI model to classify user content and transform their perceptions into analytical data.
3.1 Sampling methodology and agreement (Kappa)
To validate the DeepSeek-V3 sentiment classification model, a random stratified sample was extracted proportionally based on the total number of comments collected. According to , this approach is suitable for a heterogeneous population but can be divided into subgroups (strata) that are internally homogeneous with respect to a particular characteristic. In the present study, the years in which the comments were collected were defined as strata, ensuring that the temporal representativeness of the sample reflected the distribution of the original population.
The total sample size (
) was determined using the equation for finite populations, which considers the population size (
), the desired confidence level (
), the acceptable margin of error (
), and the proportion (
). According to the following
Equation 1(
):
where:
: Total sample size;
: Total population size (total number of comments extracted);
Error, standard deviation corresponding to the desired confidence level. For a 95% confidence level, = 1.96;
Estimated proportion of the characteristic of interest in the population. α was set t = 0.5, as recommended for situations where the proportion is unknown;
: Maximum acceptable margin of error. A confidence level of 0.05 (5%) was defined.
After determining the total sample size, the distribution of comments by stratum (year) was performed proportionally, that is, the number of elements selected from each stratum is directly proportional to the size of that stratum relative to the total population. Given by
Equation 2below (
):
where:
: Sample size for the stratum ;
: Population size of the stratum (total number of comments in the year );
: Total population size;
: Total sample size.
The analysis of the comment sample was divided into two stages: in the first, manual (human) coding was performed, and in the second, AI (DeepSeek-V3) classified the comments. These were organized into spreadsheets, and the analyses were conducted separately for each year (2017–2025), station (Sé and Trindade), and for each dimension analyzed, namely: (1) Overall sentiment (positive, negative, neutral); (2) Public Value (accessibility, safety, cost-effectiveness); (3) Travel Experience (physical/sensory, service quality, emotional/affective).
The text of the reviews was classified as “positive,” “negative,” or “neutral” (without emotional charge); the classifiers “Present” (if the criterion was present in the comment) and “Absent” (if the criterion was not present in the comment). Agreement between human coders and AI was measured using Cohen’s Kappa coefficient based on the results matrix that accounted for the classifications. Mathematically represented by the following
Equation 3(
):
where:
is the agreement coefficient (Kappa);
is the proportion of observed cases;
is the proportion of expected cases;
represents the maximum possible agreement beyond chance.
The ratio of observed cases to expected cases was calculated using Equations 4, 5, respectively ():
Observed Cases (
):
where:
is the total number of samples;
is the number of classification categories;
is the number of samples in which humans and AI agree on the category.
Expected Agreement (
):
where:
is the total number of samples classified into the “ “ category by the human;
is the total number of samples classified in the “ ” category by the AI.
The interpretation of Kappa values followed the scale proposed by , in which Kappa ranges are classified as: 0.81–1.00 (Almost perfect); 0.61–0.80 (Substantial); 0.41–0.60 (Moderate); 0.21–0.40 (Reasonable); 0.00–0.20 (Slight); <0.00 (Poor).
After achieving a score equal to or higher than Substantial agreement (above 0.61), the results of the manual coding were used as “ground truth” to train and validate the AI model. The performance of the final classifier was evaluated using the metrics Accuracy, Precision, Recall (sensitivity), and F1 Score (harmonic mean of precision and recall) (; ). The macro-average of the metrics was adopted, since the evaluations were multi-class (sentiment: positive, negative, and neutral) (). As shown in Equation 6 through 9, respectively:
Accuracy (:
Precision ():
Recall ():
F1-Score (
:
where:
are True Positives (AI and human agree on the positive class);
are True Negatives (AI and human agree on the negative class);
are False Positives (AI predicted positive, human classified as negative);
are False Negatives (AI predicted negative, human classified as positive).
The Jupyter Notebook tool was used to structure graphs and run the automated comment extraction program (Python language).
4 Results and discussion
4.1 Stations with the highest passenger volumes in the cities of São Paulo and Porto
The published the 2025 indicators on its transparency portal, which show that the busiest station in the capital is Sé Station. In 2025, an average of approximately 405,000 passengers passed through the station on weekdays, taking into account transfers between Lines 1 (Blue) and 3 (Red), as well as entries and exits. As shown in Figure 3, which displays the busiest stations in São Paulo along with the year-over-year change.
FIGURE 3
It can be observed that Sé Station, combining its two lines, accounts for approximately 33.5% of the total passengers at the 10 busiest stations listed in Figure 3. That is, 1 in every 3 passengers counted in this ranking passed through Sé Station.
Based on the most recent official data from , the station with the highest number of passengers and validations is Trindade Station. In 2025, Metro do Porto reached a historic record of 94.5 million validations, representing a 5.4% increase compared to 2024, when the figures reached 89 million. Trindade Station is considered the network’s main hub, surpassing the mark of 17.5 million annual validations—approximately 69,106 validations on average on weekdays (considering 253 weekdays in 2025)—and accounting for 18% of the entire network’s passenger traffic. Figure 4 shows the annual number of validations at Porto’s 10 busiest stations during the year 2025.
FIGURE 4
According to the data in Figure 4, Trindade station stands out significantly as the station with the highest daily ridership, surpassing all others. The Campanhã and Casa da Música stations also have a high passenger volume throughout the year, reflecting the importance of their locations as part of the network’s strategy. Therefore, Table 2 shows that the largest stations in each city are: Sé Station in São Paulo and Trindade Station in Porto.
TABLE 2
| City | Station | Line | Quantity | Average |
|---|---|---|---|---|
| São Paulo | Sé | 1 and 3 | 405,000 | Average on weekdays |
| Porto | Trindade | A, B, C, D, E, and F | 17,483,636 | Annual average of validations |
Average subway ridership according to the metric adopted by each city.
Source: Prepared by the authors, based on reports from the Porto and São Paulo Metro system (2025).
It is important to note that all results reported below derive from user-generated textual perception rather than direct physical verification of station conditions; references to cleanliness, safety, or equipment performance therefore describe how users perceive the attributes, not an independently audited physical state.
4.2 Results of the agreement analysis for Sé Station in São Paulo (Brazil)
A total of 2,196 reviews were extracted from ; after data processing and structuring, comments lacking analytical information (empty fields, isolated emojis, or random words) were excluded. The sample represents 15% of the total comments (331 comments), distributed proportionally and selected randomly for each year analyzed (according to Equations 1, 2), covering the period from September 2017 to September 2025 for the analysis of sentiment agreement, public value, and travel experience. The results of the stratified sampling are shown in Table 3.
TABLE 3
| Year | Total comments | No. of samples |
|---|---|---|
| 2017 | 179 | 27 |
| 2018 | 342 | 51 |
| 2019 | 486 | 73 |
| 2020 | 204 | 31 |
| 2021 | 197 | 30 |
| 2022 | 233 | 35 |
| 2023 | 211 | 32 |
| 2024 | 272 | 41 |
| 2025 | 72 | 11 |
| Total | 2,196 | 331 |
Results of comment stratification - Sé Station (Overall average of the analyzed dimensions).
Source: Prepared and calculated by the authors, based on and comments extracted from .
Bold values indicate the total number of comments extracted and the total sample extracted and obtained for Sé Station, São Paulo.
Kappa agreement calculations (Equations 3–5) were performed using the contingency matrix created to summarize the results obtained for each analyzed category and subcategory (sentiments, public value, and travel experience). As shown in Table 4, which presents the average results by analyzed dimension.
TABLE 4
| Dimension | Subdimension | Observed agreement average | Expected average | Average Kappa | Interpretation |
|---|---|---|---|---|---|
| General | Overall | 0.92 | 0.41 | 0.87 | Almost perfect |
| Public value | Accessibility | 0.94 | 0.63 | 0.84 | Almost perfect |
| Physical and sensory experience | 0.94 | 0.57 | 0.86 | Almost perfect | |
| Service quality (designed space) | 0.95 | 0.74 | 0.78 | Substantial | |
| Emotional/Affective experience (lived space) | 0.95 | 0.52 | 0.90 | Almost perfect | |
| Security | 0.94 | 0.59 | 0.86 | Almost perfect | |
| Value for money | 0.95 | 0.65 | 0.84 | Almost perfect | |
| Overall average | 0.94 | 0.59 | 0.85 | Almost perfect | |
Summary by subdivision of Sé station in São Paulo (2017–2025).
Source: Prepared by the authors, based on comments extracted from , , and .
Bold values indicate the overall average obtained for each column.
Table 4 shows that at Sé Station, the average Kappa coefficient for the period from 2017 to 2025 was 0.85, classified as “almost perfect,” indicating high agreement between manual and automatic classification. Over the years, a variation in the coefficient can be identified, which may result from greater complexity or ambiguity in the comments or in the distribution of classes over the years. The subcategories with the highest Kappa values were analyzed in Service Quality and Emotional/Affective Experience; compared to General Sentiment and Accessibility, these present lower values (k = 0.88 and 0.86, respectively), but still within the “near perfect” category.
Additionally, the metrics of Accuracy, Precision, Recall, and F1-Score (Equations 6–9) were calculated; the results of this analysis are shown in Table 5.
TABLE 5
| Dimension | Sub-dimension | Accuracy average | Precision average | Recall medium | F1-score average |
|---|---|---|---|---|---|
| Sentiment | Overall | 0.92 | 0.92 | 0.92 | 0.92 |
| Public value | Accessibility | 0.94 | 0.94 | 0.94 | 0.94 |
| Physical and sensory experience | 0.94 | 0.94 | 0.94 | 0.94 | |
| Service quality (designed space) | 0.95 | 0.95 | 0.95 | 0.95 | |
| Emotional/Affective experience (lived space) | 0.95 | 0.95 | 0.95 | 0.95 | |
| Security | 0.94 | 0.94 | 0.94 | 0.94 | |
| Cost-effectiveness | 0.94 | 0.94 | 0.94 | 0.94 | |
| Overall average | 0.94 | 0.94 | 0.94 | 0.94 | |
Summary by subdivision of the Sé station in São Paulo (2017–2025).
Source: Prepared by the authors, based on comments extracted from and .
Bold values indicate the overall average obtained for each column.
The average classifier performed excellently for Sé Station, with an overall average F1-Score of 0.94, ranging from 0.92 (for General Sentiment) to 0.95 (Service Quality and Emotional/Affective). Table 4 shows that the AI (DeepSeek-V3) demonstrated a good ability to identify the general sentiment (positive, negative, and neutral) expressed by users, as well as to identify mentions related to Public Value and Travel Experience, confirming that the classifier is reliable for analyzing comments.
4.2.1 Results of the sentiment analysis of the Sé in São Paulo (Brazil)
Users were classified based on star ratings and the presence of negative and positive terms within the pre-established criteria mentioned earlier. The sentiment analysis performed by DeepSeek-V3 showed a predominance of positive reviews, with an average rating of 4.04 stars; the average percentage of positive comments was 69.5%, while the average percentage of negative comments was 30.5%. Regarding measures of central tendency, the median star rating ranged from 3.8 to 4.3 during the analyzed period.
On average, 72.6% of users gave high ratings for criteria such as station integration and accessibility (4 or 5 stars), while 13.2% gave very low ratings for comfort, safety, and cleanliness (1 or 2 stars), which lowered the average.
Overall, the perception of users who use the station for modal transfers tends to be positive; however, users who use only the subway and experience access difficulties due to overcrowding or face safety or cleanliness issues (mainly outside the station) express severe dissatisfaction. This highlights that positive and negative evaluation trends are directly linked to the perception of the space (designed space) and the services provided by the São Paulo Metro and City Hall, which can either promote or hinder the creation of Public Value. As shown in Figure 5, the graph illustrates the evolution of the percentage of positive and negative sentiments in user comments about Sé Station between 2017 and 2025.
FIGURE 5
Figure 5 shows the trend of positive and negative mentions over the years. When relating these to external contexts, a pattern emerges in the sentiment curve, such as the socioeconomic crisis (2018–2019) and the urban crisis (2023), which correspond to troughs in positive sentiment and peaks in negative sentiment. The pandemic period (2020–2021) shows an improvement in user perception and a partial recovery in 2024–2025, which coincide with positive mentions of station policing in user perception.
Figure 6 illustrates the behavior of user ratings in relation to the historical contexts faced throughout the analyzed period, where it is possible to understand how crises or historical events influence the perception of subway service quality, with ratings serving as an urban barometer ().
FIGURE 6
Figure 6 shows contexts that correlate with the same periods and may explain the average rating results for the Sé station. From 2017 to 2019, there was a drop in the average rating from 3.7 to 3.9 stars, along with a higher rate of negative sentiment at 38.6%, and the user comments frequently referenced perceived insecurity (robberies), dirtiness, and overcrowding. This period was marked by a post-2015 economic crisis and a fiscal crisis in São Paulo, where there were budget cuts, infrastructure degradation, and increased social vulnerability (; ).
During the peak of the COVID-19 pandemic (2020–2022), due to a reduction in passenger numbers and the intensification of health measures, average ratings rose to 4.1 in 2021, with 80% positive comments and the emergence of frequent mentions of “clean” and “organized”; however, criticism regarding safety persists. In 2023, the drop in ratings is contextualized by the Cracolândia Crisis, which worsened security in downtown São Paulo—a situation directly reflected at Praça da Sé (São Paulo’s ground zero)— the station’s rating dropped to 3.9, and 34% of negative comments highlighted mentions of “homeless people,” “drug addicts,” “robberies,” “Cracolândia,” and “abandoned.” According to the , this period was marked by a 31% increase in the homeless population. By comparison, in 2024 there were investments in security and maintenance (), and the average rating reached its highest value for the entire period: 4.3 stars.
4.3 Results of the sentiment analysis of the trindade station in porto (Portugal)
Approximately 1,146 user reviews of the Trindade metro station were extracted from . After data processing, a stratified sample of 292 reviews (25%) was randomly selected, proportional to the total number of reviews per year (as per Equations 1, 2). Table 6 shows the results obtained from the sampling.
TABLE 6
| Year | Total of comments | No. of samples |
|---|---|---|
| 2017 | 30 | 8 |
| 2018 | 73 | 19 |
| 2019 | 35 | 9 |
| 2020 | 60 | 16 |
| 2021 | 47 | 12 |
| 2022 | 78 | 20 |
| 2023 | 243 | 62 |
| 2024 | 330 | 83 |
| 2025 | 250 | 63 |
| Total | 1,146 | 292 |
Results of the stratification of comments - Trindade Station (Overall average of the analyzed dimensions).
Source: Prepared and calculated by the authors, according to and comments extracted from .
Bold values indicate the total number of comments extracted and the total sample extracted and obtained for Trindade Station, Porto.
Additionally, the Kappa coefficient was calculated to analyze the agreement of the results (AI vs. Human) over the period and according to the dimension established by the analysis criteria. As shown in Table 7, which presents the results calculated using Equation 3 through 5 ().
TABLE 7
| Dimension | Subdimension | Average observed agreement | Expected expected average | Average Kappa | Interpretation |
|---|---|---|---|---|---|
| General | Overall | 0.92 | 0.44 | 0.85 | Almost perfect |
| Public value | Accessibility | 0.95 | 0.56 | 0.90 | Nearly perfect |
| Physical and sensory experience | 0.96 | 0.57 | 0.91 | Almost perfect | |
| Service quality (designed space) | 0.97 | 0.70 | 0.89 | Almost perfect | |
| Emotional/Affective experience (experienced space) | 0.94 | 0.29 | 0.90 | Almost perfect | |
| Security | 0.98 | 0.83 | 0.83 | Almost perfect | |
| Value for money | 0.97 | 0.65 | 0.92 | Almost perfect | |
| Overall average | 0.96 | 0.58 | 0.89 | Almost perfect | |
Summary by subdivision of the Trindade station in Porto (2017–2025).
Source: Prepared by the authors, based on comments extracted from , , and .
Bold values indicate the overall average of the values obtained for each column.
The Kappa coefficient for Trindade Station (2017–2025) of 0.89 (“almost perfect”) indicates that there is agreement between the manual classification and the AI-based classification; that is, the AI classification was better than what would be expected by chance. The highest Kappa values were observed for safety, cost-benefit, and emotional/affective experience (k = 0.92). Table 8 shows the results of the Accuracy, Precision, Recall, and F1-Score calculations (Equation 6 through 9).
TABLE 8
| Dimension | Subdimension | Accuracy average | Precision average | Recall medium | F1-score average |
|---|---|---|---|---|---|
| Sentiment | Overall | 0.93 | 0.93 | 0.93 | 0.93 |
| Public value | Accessibility | 0.94 | 0.94 | 0.94 | 0.94 |
| Physical and sensory experience | 0.96 | 0.94 | 0.96 | 0.96 | |
| Service quality (designed space) | 0.96 | 0.96 | 0.96 | 0.96 | |
| Emotional/Affective experience (lived space) | 0.95 | 0.95 | 0.95 | 0.95 | |
| Security | 0.98 | 0.98 | 0.98 | 0.98 | |
| Cost-effectiveness | 0.96 | 0.96 | 0.96 | 0.96 | |
| Overall average | 0.95 | 0.95 | 0.95 | 0.95 | |
Summary by subdivision of the Trindade station in Porto (2017–2025).
Source: Prepared by the authors, based on comments extracted from and .
Bold values indicate the overall average of the values obtained for each column.
The automatic classifier performed excellently for Trindade Station with an overall average F1-Score of 0.95, ranging from 0.93 (Overall Sentiment) to 0.98 (Safety). DeepSeek-V3 demonstrated a high ability to correctly identify the sentiment expressed in the comments with recall values of 0.93, indicating a low occurrence of false positives and false negatives. The sub-dimension with the highest F1 score was safety, suggesting that the textual expressions refer to the service or a sense of safety. The overall average for all metrics confirms that the automatic classifier is reliable.
4.3.1 Results of the sentiment analysis of trindade in porto (Portugal)
The sentiment analysis of the 1,146 comments, collected between 2017 and 2025 on Google Maps, reveals that user perception of the service is predominantly positive. The median rating remained consistently at 5 stars throughout the entire period, and the average rating never fell below 4.10, indicating consistent satisfaction. As shown in Figure 7. However, the volume of comments grew exponentially, especially starting in 2023, reflecting user engagement with the review platform.
FIGURE 7
It is worth noting that 2021 stood out as a peak year for excellence, with an average rating of 4.67 stars and 92.2% positive reviews, with no negative comments (1- and 2-star ratings), suggesting a post-pandemic impact, where the resumption of services following social isolation (and restrictive measures) may have been significantly valued by users. However, in subsequent years, particularly in 2024 and 2025, the frequency of positive mentions returns to pre-2021 levels, with an increase in criticism. From 2022 to 2025, sentiment analysis highlights terms such as “crowded,” “noise,” and “tourists,” indicating that the rise in tourism is reflected in increased passenger flow on the subway, which ultimately strains the infrastructure and negatively impacts user comfort and experience. Figure 8 illustrates significant passenger crowding on the platform at Trindade Station.
FIGURE 8
Figure 9 shows the evolution of the average rating over the years and events that contextualize user perception of metro use.
FIGURE 9
As shown in Figure 10, events can be identified that correlate with the rating results. During the Tourism Expansion period (2017–2019), the Port saw an increase in the number of tourists, thereby boosting usage of the transit network (). The peak in satisfaction recorded in 2021 (4.63 stars) aligns with the post-pandemic recovery phase, during which, despite partial lockdowns, there was a 5.9% increase in demand for transportation compared to 2020, highlighting the perception of a “return to normalcy” (; ). However, starting in 2023, a return to dissatisfaction is observed, coinciding with the historic record of 6 million overnight stays in the city, which is associated with the saturation of the central infrastructure (; ).
FIGURE 10
4.4 Discussion and analysis of user perception of public value and travel experience at sé and trindade stations
Sentiment analysis suggests that the primary driver of Public Value is users’ recognition of Modal Integration as efficient; this criterion consistently remained above 85% positive ratings. Comments highlight the connections between multiple subway and modal lines, acknowledging their structural role in the city (São Paulo’s ground zero). Based on this, the station is also perceived as a public facility that serves a social function, enabling access to jobs, healthcare, education, and culture for millions of people, especially the low-income population that relies more heavily on public transportation.
Similarly, Trindade Station is mentioned positively by users, with its modal integration highlighted as a “central hub” of the Metro do Porto, connecting all lines and offering intermodality (buses, trains, and the airport), which aligns directly with sustainable mobility policies perceived as a collective benefit, reinforcing the notion that infrastructure serves a broad public interest (). Its location is also highlighted as “central” and “accessible to all areas”; users perceive that this facilitates travel and revitalizes the city (“spillover effect”), as discussed by . In years when maintenance and safety are positively evaluated (such as in 2021 and 2022), users associate the station with an “organized,” “efficient” environment, that is, one that fosters wellbeing and public care, which contributes to the generation of public value.
However, the analysis of the Travel Experience, understood as the practical and subjective experience of travel (), and its layers (physical experience, service quality, and emotional/affective experience), reveals persistent critical issues for Trindade Station, such as the confusing interface of the ticket vending machines or counter service considered unempathetic or insufficient, which are described as the main source of frustration throughout nearly all the years analyzed. Figure 10 shows passenger queues waiting to use the ticket vending machines at Trindade Station by the authors.
Sé Station, on the other hand, stands out for the concentration of the most severe and emotional criticisms, reflecting the user’s direct contact with the space, where Security (72% negative) and Cleanliness (68% negative) are the most poorly rated categories. The comments of users describe the station with negatively charged language (“smelly,” “scary,” “disgusting”), indicating a stressful experience that generates fear and revulsion, devaluing the travel experience and the public value of transportation, making it less attractive. As the main line in São Paulo, there are also issues related to overcrowding and a perception of insecurity that directly compromise comfort, dignity, and the sense of wellbeing during the journey. The experience ceases to be neutral or functional and becomes actively unpleasant. The discrepancy between the Median (4–5 stars) and Average (3.8–4.3 stars) reveals that, while many have an “acceptable experience,” a significant minority have extremely bad experiences and end up rating it with one star, pulling the average down.
The data from Sé illustrates the relationship proposed by , where public value is called into question when public infrastructure fails to provide a minimum level of dignity—in this case, the valuation or devaluation of service use—as a relationship between the individual and society. In 2023, there was a spike in comments referring to “Cracolândia” (the largest hub for drug addicts and socially vulnerable individuals in the state of São Paulo) and “robberies,” which is associated with a measurable drop in the overall rating; this demonstrates that when there is external insecurity and overcrowding (a very negative collective experience), it impacts the legitimacy of the service and consequently the public policy.
It is important to distinguish two categories of negative drivers identified in this analysis. At Sé Station, negative sentiment is largely driven by broader urban-management and public-safety conditions surrounding the station (the Cracolândia crisis, homelessness, perceived insecurity), which fall under municipal and public-safety governance rather than the metro operator’s direct control. At Trindade Station, by contrast, negative sentiment concentrates on rail-operation factors (ticketing-machine performance, congestion management), which are within the operator’s direct remit. Collapsing both into a single ‘metro facility’ explanation would misattribute responsibility and weaken the precision of the policy implications that follow.
4.4.1 Analytical cross-tabulation of users’ perceptions of public value and travel experience at both stations
Table 9 shows the average distribution of comments according to combinations of Public Value (PV) and Travel Experience (TE) during the analyzed period. The values were estimated based on the positivity rates by category.
TABLE 9
| PV x TE combination | Sé station (%) | Trindade station (%) |
|---|---|---|
| Positive PV + positive TE | 38 | 52 |
| Positive VP + Negative EV | 42 | 18 |
| Negative VP + positive EV | 5 | 12 |
| Negative VP + Negative EV | 15 | 18 |
| Total | 100 | 100 |
Distribution of comments across the Public Value and Travel Experience dimensions for the Sé and Trindade stations.
Source: Prepared by the authors based on DeepSeek-V3, sentiment analysis.
Bold values indicate the total percentage distribution of positive and negative comments (which sum to 100%).
The data in Table 9 demonstrate a pattern of dissociation, as Sé users separate the collective dimension (the station’s public value as an essential hub) from the individual dimension (travel discomfort), allowing the service’s legitimacy to persist despite the negative experience (dissatisfaction with travel comfort or safety in the surrounding area and inside the station). This dissociation is greater where dependence on the service is most necessary, suggesting adaptation to precarious conditions. This explains why 42% of the comments combine positive Public Value with a negative Travel Experience (more than double the 18% observed in Trindade). This result confirms proposition that the Public Value of a service can sustain its collective legitimacy.
As for Trindade Station, Table 9 reveals a distinct pattern: only 18% of comments show positive Public Value with a negative Travel Experience, while 52% simultaneously record positive ratings in both dimensions. This result indicates that, in the context of Trindade, these dimensions are perceived in a more integrated manner; when the experience deteriorates, the assessment of public value tends to follow suit. This behavior is explained by the concentration of tourist traffic at the station, where the location is not merely a daily transit point but a primary point of contact with the city’s mobility system, a factor that directly shapes the perception of public value based on the immediate experience (ticket machine functionality, clarity of signage, customer service). This is also demonstrated in Table 10 below, where service quality appears in 88% of 1–2 star ratings, constituting the main bottleneck for system satisfaction.
TABLE 10
| Subdimension of the travel experience | Sé station (%) (impact of 1–2 star ratings) | Trindade station (%) (impact of 1–2 star ratings) |
|---|---|---|
| Physical/sensory (cleanliness, comfort, crowding, safety) | 78 | 28 |
| Service quality (machines, customer service, signage) | 48 | 88 |
| Emotional/affective (frustration, fear, overall satisfaction) | 62 | 52 |
Distribution of 1–2 stars by subdimension of the Travel Experience (TE).
Source: Prepared by the authors based on the sentiment analysis from DeepSeek-V3.
Table 10 shows the proportion of 1–2 star ratings associated with criticisms of the Travel Experience (subdimension cited as a problem).
Table 10 shows which subdimension of the Travel Experience has the greatest influence on the lowest ratings, demonstrating a distinction between the experiences at the two stations. At Sé Station, the physical and sensory dimension (which encompasses perceptions of cleanliness, comfort, crowding, and safety) is present in 78% of ratings with scores of 1–2, confirming the deterioration of the physical and sensory experience of the perceived space, which is the primary driver of dissatisfaction. This result aligns with literature on the “conceived space” as the primary layer of the travel experience; that is, when passengers are exposed to adverse conditions such as overcrowding, insecurity, or dirt, their evaluation collapses.
However, investment practices differ between the two systems. For Sé, urban planning, public safety, and access control interventions yield the greatest return in user satisfaction. For Trindade, improvements to ticketing interfaces and training for customer service staff are key drivers of better experiences.
In the context of Public Value, Table 11 shows the proportion of positive comments regarding the sub-dimensions of Public Value that sustain positive sentiment during years of overall negative sentiment. At Sé station, these were the years 2018–2019 and 2023 (major crisis years), and at Trindade, the years 2023–2025 (years of high tourist demand).
TABLE 11
| Public value subdimension | Sé station (% positive comments) | Trindade station (% positive comments) |
|---|---|---|
| Modal integration (connections between lines, intermodality) | 82% | 88% |
| Accessibility and location (centrality, ramps, elevators) | 65% | 70% |
| Strategic utility (hub, time savings, urban landmark) | 71% | 75% |
Distribution of positive comments on the sub-dimensions of Public Value (PV).
Source: Prepared by the authors based on DeepSeek-V3, sentiment analysis.
Table 11 highlights the possible factors that underpin the Public Value of transportation in crisis contexts. As previously discussed, modal integration is the most resilient sub-dimension of Public Value in both systems, maintaining high levels of positive perception even during periods of greater deterioration in the Travel Experience: 82% in Sé during the crisis years and 88% in Trindade during the years of highest tourist pressure on the system. This conclusion has relevant theoretical implications for the operationalization of Public Value in , where the perception of the service’s collective utility sustains systemic legitimacy even when other dimensions are under pressure.
The recommendations below are strategic priorities derived from user-perceived issues; their engineering feasibility (platform, layout, signage placement, ticketing hardware upgrades) would require validation through dedicated technical studies with the responsible transport operators, which falls outside the scope of this UGC-based analysis.
From a public policy perspective, this result suggests investments in modal integration and connectivity (line expansion, improved intermodal signage, fare integration) as they yield greater returns in terms of perceived legitimacy compared to other dimensions. This pattern aligns with the findings of regarding the appropriateness of cost-benefit analysis in assessing public value in transportation, where collective perception is not reduced to individual expectations, and policies that strengthen the systemic function of public transportation tend to be more politically sustainable even when the operational quality of the service varies. As shown in Figure 11, which illustrates the temporal evolution of the dissociation between Public Value (PV) and Travel Experience (TE) from 2017 to 2025.
FIGURE 11
Figure 11 illustrates the different trajectories of the two dimensions analyzed at Sé Station over the eight-year observation period. While positive Public Value remained stable, fluctuating between 68% and 76%, positive Travel Experience showed greater variation, peaking in 2021 (68%), which correlates with the post-pandemic period of lower passenger flow and greater social distancing, and dropping to 43% in 2023, when the intensification of insecurity incidents in the Cracolândia region generated negative sentiment regarding passengers’ sensory and emotional experience.
The difference between the two lines in 2023 (with a 29-percentage-point gap between positive Public Value (71%) and positive Travel Experience (42%)) is the most relevant data point in the graph, demonstrating that even at the lowest point of experience recorded during the period, the majority of users who criticized the station still recognized its collective utility. This pattern empirically validates the dissociation hypothesis and confirms that Public Value and Travel Experience are not redundant, as they measure genuinely distinct dimensions of the perception of public service.
At Trindade station, however, Figure 11 shows a different pattern: the positive Public Value and positive Travel Experience lines move more closely together over the period, with occasional divergences during years of higher operational pressure. In this case, the point of greatest divergence occurs in 2021, when positive PV reached 88% while positive TE reached 84% (both on the rise), reflecting the pandemic period with lower tourist flow.
It should be noted that this temporal alignment is correlational: while external events (economic crisis, pandemic, tourism surge) coincide with shifts in sentiment, an alternative explanation cannot be ruled out, namely, that gradual infrastructure aging is itself the primary driver of declining experience, with external events acting as amplifiers of an already-deteriorating baseline rather than as independent causes. Distinguishing between these hypotheses would require quantitative control variables (maintenance investment data, crime statistics) that fall outside the scope of this UGC-based study.
5 Conclusion
Based on the above, the results show that, even in different socioeconomic and demographic contexts, the two cities play an important role as centers of integration and are viewed by users as having a structural and social function in urban mobility. When comparing the results of the sentiment analysis between the stations, it is evident that at Sé Station, the perception of Public Value appears in a collective form, highlighting the efficiency of modal integration even during crises in the user’s travel experience.
As for Trindade Station, high passenger satisfaction rates are observed, evidenced by consistent positive mentions over the years, highlighting cleanliness and intermodality with access to buses, trains, and the airport. However, the study found that due to the city of Porto’s success as a tourist destination—which, according to , is the second most visited municipality in Portugal—new pressures have emerged on the system, such as overcrowding and glitches in the Andante ticketing system, which alter users’ perceptions of service quality. A comparison between stations shows that improvements to the travel experience (such as policing and maintenance) have visible effects for users, thereby increasing wellbeing and the legitimacy of the system, adding Public Value.
The findings of this study are consistent with recent international literature on sentiment analysis in transportation by , who demonstrated the superiority of CGU in capturing qualitative and quantitative dimensions of mobility in Portugal, as well as the authors , who identify similar patterns of criticism regarding maintenance and accessibility in Chinese Google Maps; the results presented here confirm that the reviews are capable of capturing external systemic shocks, functioning as an “urban thermometer” as state. Additionally, emphasize that Latin American systems face challenges, particularly in the areas of safety and cost-effectiveness, which are structurally distinct from the problems of the European system, justifying the need for models tailored to each context that are structurally distinct from European ones.
Consequently, by identifying the strengths and weaknesses of the metro system’s stations from the user’s perspective, the needs perceived by users are identified, thereby contributing to the alignment of metro transportation with the UN Sustainable Development Goals (SDGs), specifically targets 9.1 (development of quality, sustainable, reliable, and resilient infrastructure) and 11.2 (providing access to safe and sustainable transport systems for all). However, the study is limited by: (i) the data come exclusively from Google Maps reviewers, and likely underrepresent local residents with infrequent online-review habits, low-income users or elderly passengers; (ii) the inability to make a direct comparison between São Paulo and Porto (due to differences in population, cultural assessment, etc.); (iii) the identified temporal associations being correlational rather than causal; (iv) visual verification in this study was partial: a targeted site visit to Trindade Station allowed direct photographic documentation of overcrowding and ticketing, and illustrative user-generated images from Google Maps further evidence overcrowding at Sé Station; however, no equivalent visual verification of security- and cleanliness-related claims at Sé Station was possible, nor were architectural plans, transfer-flow diagrams, or accessibility maps obtained for either station; (v) the total review volume differs substantially between stations (2,196 for Sé vs. 1,146 for Trindade) and is unevenly distributed across years, with a larger Sé sample concentrated in 2017–2019 and a marked increase in Trindade reviews after 2023, which was only partially mitigated by using percentage-based rather than absolute comparisons; no year-weighting correction was applied, so cross-year and cross-station percentages should be interpreted with this imbalance in mind; (vi) the keyword framework (Table 1) was developed in general Portuguese and was not subjected to a dedicated ablation for São Paulo colloquialisms or Porto-specific/tourist dialectal expressions; dialectal divergence between the two cities may introduce differential misclassification, which was not directly tested, although the high inter-rater Kappa scores obtained for both stations (Tables 7, 8) offer indirect evidence of overall classifier reliability.
For Sé Station, we propose three monitorable, UGC-derived proxy indicators to track progress: (i) frequency of safety/insecurity mentions in the station’s immediate surroundings, tracked quarterly; (ii) share of negative reviews explicitly referencing night-time conditions, where time-stamp data permit; and (iii) frequency of ‘old/worn/outdated’ mentions regarding platforms, tracked annually as a proxy for perceived infrastructure renewal needs. For Trindade Station, we propose: (iv) the monthly share of ticketing-related complaints, benchmarked against tourist-season peaks, to guide seasonal staffing and signage measures; and (v) the frequency of language-related confusion mentions, to guide prioritization of multilingual ticketing-interface upgrades. These proxies allow the same UGC methodology used in this study to serve as a low-cost monitoring tool, pending formal engineering assessments.
Nevertheless, the findings offer a method applicable to sentiment analysis in transportation. Therefore, it can be concluded that the study’s results introduce the notion that Public Value is not perceived uniformly by users, since in times of crisis the system’s legitimacy may be called into question even when operational efficiency is maintained, due to the fact that perceptions of the travel experience are affected. They also allow for the formulation of propositions for future research, which stem from the analysis of longitudinal sentiments and the comparison between stations, without a confirmatory nature, given the exploratory nature of the research. The first is that the perception of Public Value, measured via CGU, is more sensitive to external public safety events than to internal operational failures, especially in high-demand systems in Latin America (such as Sé Station); the second would be that in systems with a high structural satisfaction index (such as Trindade Station), the Travel Experience is primarily affected by interface and sensory comfort factors rather than by accessibility or safety failures. Finally, the third proposition is that perceived Modal Integration is the primary generator of Public Value in both contexts, regardless of socioeconomic and demographic differences among the analyzed systems.
These findings also connect with goals already underway in the operators’ most recent institutional reports. Metrô de São Paulo’s Integrated Report 2025 highlights modal integration and the modernization of digital ticketing systems as strategic company priorities—which directly converges with this study’s third proposition, that perceived Modal Integration is the primary generator of Public Value in both analyzed contexts, and suggests that the operator’s already-planned investments have the potential to act precisely on the dimension users value most (). Metro do Porto’s Annual Report 2025, in turn, details a new expansion phase for the network (including the Dragão-Souto, Maia II, and São Mamede axes and the connection to the ISMAI-Trofa axis, which together could add up to 42 new stations and approximately 37 km to the network), which may, in the medium term, redistribute the passenger and tourist flow currently concentrated at Trindade Station, mitigating part of the ticketing-system overload identified in this study (). Although these institutional plans were not designed from the UGC-based perceptions analyzed here, this convergence suggests that this study’s findings can serve as a complementary, user-perception-based indicator for monitoring the effectiveness of these already-planned initiatives.
Additionally, the study contributes in two distinct ways to the literature on transportation policy and public management. First, theoretically, by operationalizing the concepts of Public Value and Travel Experience based on User-Generated Content (UGC), demonstrating that the three central dimensions of social legitimacy (accessibility), political authorization (safety), and operational feasibility (cost-benefit), as well as the dimensions of conceived and lived space and service quality, can be reliably extracted from spontaneous reviews on Google Maps. Finally, empirically, through the presentation of longitudinal evidence (2017–2025) from two metro stations at different stages of maturity (São Paulo, 51 years old and 11.4 million inhabitants; and Porto, 23 years old and 252,000 inhabitants), we demonstrate how external shocks (pandemics, security crises, surges in tourism) affect user perception at central stations. These contributions bridge the theory of Public Value with the user experience in transportation through citizen-generated data.
Statements
Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
Author contributions
HS: Data curation, Investigation, Software, Writing – original draft, Writing – review and editing. ÁC: Conceptualization, Methodology, Supervision, Validation, Writing – review and editing. RO: Supervision, Validation, Writing – review and editing.
Funding
The author(s) declared that financial support was received for this work and/or its publication. Faculty of Engineering, University of Porto, and São Paulo State Research Support Foundation (FAPESP), Overseas Research Internship Grant (BEPE), grant number 2025/28625-1.
Acknowledgments
The authors would like to thank the São Paulo Research Foundation (FAPESP) for financial support, process number 2025/28625–1. We also express our gratitude to the Federal Institute of Education, Science and Technology of São Paulo - Campus São Paulo (IFSP-SPO) and the Faculty of Engineering of the University of Porto (FEUP) for their institutional support and contributions to this work.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was used in the creation of this manuscript. The authors declare and assume full responsibility for the use of generative AI in the preparation of this manuscript. Generative AI was used to: (1) Perform Sentiment Analysis on User-Generated Content (UGC) collected from Google Maps using the DeepSeek-V3 model, classifying comments according to overall sentiment (positive, negative, or neutral) and the sub-dimensions of Public Value and Travel Experience, based on criteria and prompts developed and validated by the authors themselves; (2) Assist in structuring and organizing tables and graphs based on the quantitative results obtained (frequencies, percentages, co-occurrence matrices, and dissociation matrices), using data previously processed and validated by the authors. In both cases, the authors fully supervised the process, defining the methodological criteria, classification parameters, and prompts used, in addition to validating the results generated by AI through stratified sampling and comparison with human coding (Cohen’s Kappa coefficient). Generative AI was not used for writing the scientific text, generating original data, formulating hypotheses, or providing theoretical interpretations of the results; these steps were conducted exclusively by the authors. The authors reaffirm their full responsibility for the accuracy, integrity, and ethical appropriateness of all content presented in this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
References
1
AhmadM.BatyrshinI.SidorovG. (2025). Sentiment analysis using a large language model–based approach to detect opioids mixed with other substances Via social media: method development and validation. JMIR Infodemiology5, 70525. 10.2196/70525
2
AIMA (2024). Agency for migration and asylum integration. Available online at: https://aima.gov.pt/media/pages/documents/a109224112-1726562672/rma-2023.pdf (Accessed December 11, 2025).
3
AmorimA. M. M. C.GonçalvesL. A. G. T.IsodaM. K. T. (2022). “The São Paulo metro in Design: architecture and metropolis,”. São Paulo. 10.11606/9786589514251
4
AMP (2023). “Metro do Porto,”. Porto. Available online at: https://www.amp.pt/item/relacoes-institucionais-or-metro-do-porto/ (Accessed June 22, 2026).
5
BalzaL. H.Gómez-ParraN.PasmanC.SerebriskyT.SosaB. (2023). “Urban Voices: exploring perceptions of infrastructure services in Latin American megacities,”. Inter-American Development Bank. 10.18235/0005358
6
Brazilian Forum on Public Safety – FBSP. (2024). Brazilian yearbook of public safety. Available online at: https://forumseguranca.org.br/wp-content/uploads/2024/07/anuario-2024.pdf (Accessed April 30, 2026).
7
Brazilian Institute of Geography and Statistics - IBGE (2022). São Paulo. Available online at: https://www.ibge.gov.br/cidades-e-estados/sp/sao-paulo.html. (Accessed April 28, 2026).
8
BussabW. O.MorettinP. A. (2017). Basic Statistics. 9th ed.São Paulo, Brazil: Saraiva Educação.
9
CaldeiraM.RaskovaE.VázquezI. B.CostaA. (2024). The adequacy of cost-benefit analysis in the assessment of public value: a case study from the transportation sector. Adm. and Soc.56, 886–906. 10.1177/00953997241260408
10
CamposL. P. (2025). in Dissertation [master’s Thesis in Urban Planning and Design] – Faculty of Engineering and Faculty of Architecture, University of Porto (Porto), 100.
11
CohenJ. A. (1960). Coefficient of agreement for nominal scales. N. Y. Univ. Educ. Psychol. Meas.20 (1), 37–46. 10.1177/001316446002000104
12
CostaH. S. (2025). Automated System for Collecting Reviews on Google Maps BComputer program. São Paulo. Available online at: https://github.com/Henrique-Sc/Google-Maps-Comment-Collector (Accessed April 17, 2026).
13
CrosbyB. C.BloombergL. (2015). Creating Public Value in practice: Advancing the Common Good in a multi-sector, shared-power, no-one-wholly-in-charge World. New York: Routledge, 12–18.
14
DeepSeek (2026). DeepSeek-V3 [large language model - LLM]. Available online at: https://chat.deepseek.com/ (Accessed May 7, 2026).
15
FerreiraC. M. C.GonçalvesM. B. (2020). The expansion of the São Paulo metro and the challenges of urban mobility in the Brazilian metropolis. Rev. dos Transp. Públicos – ANTP42 (164), 55–78. Available online at: https://xviiienanpur.anpur.org.br/anaisadmin/capapdf.php?reqid=330 (Accessed April 30, 2026).
16
Getulio Vargas Foundation - FGV (2024). Challenges of urban mobility in Brazil: challenges and perspectives. Available online at: https://transportes.fgv.br/sites/transportes.fgv.br/files/revista_mt.pdf (Accessed April 30, 2026).
17
Google Maps (2026a). Sé Station. São Paulo, Brazil. Available online at: https://maps.app.goo.gl/GNeKof2RsW3ZpjUT7 (Accessed March 23, 2026).
18
Google Maps (2026b). Trindade Station. Porto, Portugal. Available online at: https://maps.app.goo.gl/MXdGwUupxQrgKqhq7 (Accessed March 23, 2026).
19
Government of the State of São Paulo (2026). Network map. Available online at: https://www.metro.sp.gov.br/wp-content/uploads/2026/03/mapa-de-rede.pdf (Accessed April 9, 2026).
20
HasibK. M.NaseemU.KeyaA. J.MaitraS.MithuK.AlamM. G. R. (2023). A systematic review on airlines industries based on sentiment analysis and topic modeling. Res. Square1, 1–41. 10.21203/rs.3.rs-3475984/v1
21
Institute of Applied Economic Research - IPCA (2020). Territorial policies in times of multiple crises: challenges and prospects for Brazil in the 2020s. Available online at: https://repositorio.ipea.gov.br/server/api/core/bitstreams/4c58f47f-bb2c-4800-9c27-b8d1032394b2/content (Accessed February 2, 2026).
22
IPCC (2021). Climate Change 2021: the Physical Science Basis. Cambridge University Press. Available online at: https://www.ipcc.ch/report/ar6/wg1/downloads/report/IPCC_AR6_WGI_FullReport.pdf (Accessed February 2, 2026).
23
JanuárioJ. F.CostaA.Oliveira CruzC.Miranda SarmentoJ.Faria e SousaV. (2021). Transport infrastructure, accessibility, and spillover effects: an empirical analysis of the Portuguese real estate market from 2000 to 2018. J. Res. Transp. Econ.90, 101130. 10.1016/j.retrec.2021.101130
24
LandisJ. R.KochG. G. (1977). The measurement of observer agreement for categorical data. Biometrics33 (1), 159–174. 10.2307/2529310
25
LyrioB. (2025). “Trajectories of Brazilians in Lisbon: intersections between lifestyle migration and international student mobility,” in REMHU: Interdisciplinary Journal of Human Mobility (Brasília, DF), 33. Available online at: https://www.scielo.br/j/remhu/a/4qv5HffJ6m7VxtZLpdLZNzh/?lang=pt (Accessed December 12, 2025).
26
MarchesiR. (2019). “The subway isn’t just for big cities, as shown by the city of Porto, Portugal,” in National Association of Rail Passenger Transporters - ANP Trilhos, Articles (Porto). Available online at: https://anptrilhos.org.br/metro-nao-e-apenas-para-cidade-grande-como-mostra-a-cidade-do-porto-em-portugal/ (Accessed February 1, 2026).
27
Metro do Porto (2022). 2021 Annual Report and Financial Statements. Porto: Metro do Porto, S.A, 1645–1694. Available online at: https://www.metrodoporto.pt/metrodoporto/uploads/document/file/598/metrodoporto_r_c_2021.pdf (Accessed May 5, 2026).
28
Metro do Porto (2025a). History. Available online at: https://www.metrodoporto.pt/pages/321 (Accessed February 1, 2026).
29
Metro do Porto (2025b). Metro at maximum levels of satisfaction, loyalty, and image. Available online at: https://en.metrodoporto.pt/pages/420?news_id=574 (Accessed November 25, 2025).
30
Metro do Porto (2025c). Annual Report and Financial Statements 2024. Porto: Metro do Porto S.A. Available online at: https://www.metrodoporto.pt/metrodoporto/uploads/writer_file/document/674/renumeracao_os_2024.pdf (Accessed April 30, 2026).
31
Metro do Porto (2025d). Sustainability Report 2022–2024. Porto: Metro do Porto, S.A. Available online at: https://www.metrodoporto.pt/metrodoporto/uploads/document/file/748/rs_2022_2024.pdf (Accessed: April 16, 2026).
32
Metro do Porto (2025e). Annual Report and Financial Statements 2025. Porto: Metro do Porto S.A. Available online at: https://www.metrodoporto.pt/metrodoporto/uploads/document/file/797/rc_metro_do_porto_2025_digital_lowquality.pdf (Accessed July 27, 2026).
33
Metro do Porto (2026). Number of validations in 2025. Available online at: https://www.metrodoporto.pt/pages/863?news_id=606 (Accessed April 11, 2026).
34
MeynhardtT. (2009). Public value Inside: what is public value creation?Int. J. Public Adm., 32, 192–219. 10.1080/01900690902732632
35
MooreM. H. (1995). Creating Public Value: Strategic Management in Government. Harvard University Press, 70–76. Available online at: https://raggeduniversity.co.uk/wp-content/uploads/2025/03/CreatingPublicValueStrategicManagementinGovernmentbyMarkH.Moore-a-_compressed.pdf (Accessed February 3, 2026).
36
National Association of Public Transport – ANTP. (2025). Public Transport Journal. São Paulo: 47, 157, 69–86. Available online at: https://files.antp.org.br/2025/9/10/rtp157-e.pdf (Accessed May 5, 2025).
37
National Association of Rail Passenger Carriers – ANPTrilhos (2025). Overview of the Metro-Rail Sector in Brazil 2024. Brasília: ANPTrilhos, 7–28. Available online at: https://anptrilhos.org.br/wp-content/uploads/2025/03/ANPTrilhos-Balanco-do-Setor-Metroferroviario-2024.pdf (Accessed May 5, 2025).
38
National Institute of Standards and Technology (NIST); Center for AI Standards and Innovation (CAISI) (2025). Evaluation of DeepSeek AI Models. Gaithersburg: NIST/CAISI, 2025. Available online at: https://www.nist.gov/system/files/documents/2025/09/30/CAISI_Evaluation_of_DeepSeek_AI_Models.pdf (Accessed April 28, 2026).
39
National Institute of Statistics - INE (2024). Tourism statistics - 2023. Available online at: https://www.ine.pt/ngt_server/attachfileu.jsp?look_parentBoui=675610086&att_display=n&att_download=y (Accessed April 12, 2026).
40
National Transportation Confederation – CNT; National Association of Urban Transportation Companies – NTU (2024). CNT survey on urban population mobility 2024. Brasília: CNT. Available online at: https://static.poder360.com.br/2024/08/pesquisa-cnt-mobilidade-urbana.pdf (Accessed April 30, 2026).
41
OliveiraB. R.GomesS. C. C.TobiasM. S. G. (2022). Social inequality and Urban Mobility: the case of Santarém - brazil1st ed. 10.37885/220709471
42
OliveiraS. C.PelliccioniM. C. F.WestphalM. F. (2022). Urban Mobility in Times of Pandemic. São Paulo, Brazil: University of São Paulo (USP), Institutional Repository. Available online at: https://repositorio.usp.br/directbitstream/015f28ea-7902-4570-ba5e-257fc9ea9dba/Mobilidade%2BUrbana.pdf (Accessed April 23, 2026).
43
OliveiraR. R.de, AlvaradoE. F.LopesL. I. V.SilvaN. R.da, MolinariG. E.GuimarãesP. H. M. (2007). Serviços prestados pelos metrôs de Buenos Aires e São Paulo: Análise do conteúdo gerado da plataforma Tripadvisor. Revista Rosa dos Ventos - Turismo e Hospitalidade17 (1). 10.18226/21789061.v17ip170115
44
OliverR. L. (1980). A cognitive model of the antecedents and consequences of satisfaction decisions. J. Mark. Res.17 (4), 460–469. 10.1177/002224378001700405
45
OsborneS. P. (2018). From public service-dominant logic to public service logic: are public service organizations capable of co-production and value co-creation?Public Manag. Rev.20 (2), 2025–2231. 10.1080/14719037.2017.1350461
46
PORDATA (2024). Porto - tourism: portrait of municipalities. Available online at: https://retratos.pordata.pt/turismo/porto (Accessed February 25, 2026).
47
PORDATA (2025). Population - Porto 2024. Available online at: https://retratos.pordata.pt/populacao/porto (Accessed February 1, 2026).
48
RochaH.FilgueirasM.TavaresJ. P.FerreiraS. (2023). Public transport usage and perceived service quality in a large metropolitan Area: the case of Porto. Sustainability15 (7), 6287. 10.3390/su15076287
49
São Paulo Metro (2026a). Demand Indicators 2025. São Paulo, Brazil: Companhia do Metropolitano de São Paulo (Metrô). Available online at: https://transparemencia.metrosp.com.br/dataset/demanda/resource/a2a56f78-6742-4797-a6c0-ba0c8e4ec710 (Accessed May 6, 2026).
50
São Paulo Metro (2026b). Integrated Report 2025 (RI 2025). São Paulo, Brazil: Companhia do Metropolitano de São Paulo (Metrô). Available online at: https://af0020.azurewebsites.net/wp-content/uploads/2026/03/RI-2025_MetroSP_130326-3.pdf (Accessed May 6, 2026).
51
São Paulo Municipal Secretariat for Social Assistance and Development (SEDS) (2022). Dashboard on the Homeless Population and Social Vulnerability. Available online at: https://prefeitura.sp.gov.br/web/secretaria_executiva_de_projetos_estrategicos/w/painel-da-popula%C3%A7%C3%A3o-em-situa%C3%A7%C3%A3o-de-rua-e-vulnerabilidade-social (Accessed February 18, 2026).
52
SEF (2024). Foreign population residing in Portugal. SEFSTAT Statistics Portal. Available online at: https://sefstat.sef.pt/forms/distritos.aspx (Accessed December 11, 2025).
53
SokolovaM.LapalmeG. A. (2009). Systematic analysis of performance measures for classification tasks. Inf. Process. and Manag.45 (4), 427–437. 10.1016/j.ipm.2009.03.002
54
SouzaH. A. L.PintoM. F.OliveiraR. R. deMachadoS. T.CrespiM. G.TittonelM.et al (2025). Sentiment analysis of user-generated content in the evaluation of subway services: a comparative longitudinal study between Sé Station (São Paulo) and Central do Brasil Station (Rio de Janeiro). Braz. J. Prod. Eng.11 (4), 341–358. 10.47456/bjpe.v11i4.51173
55
TaherdoostH.MadanchianM. (2023). Artificial intelligence and sentiment analysis: A competitive research review. Computers12 (2), 37. 10.3390/computers12020037
56
TorresE. C. M.Picado-SantosL. G. de. (2025a). “Using sentiment analysis to study the potential for improving sustainable mobility in university campuses,”, 17. Lisbon - Portugal, 6645. 10.3390/su17146645
57
TorresE. C. M.Picado-SantosL. G. (2025b). Sentiment analysis and topic modeling in transportation. Appl. Sci.15 (12), 6576. 10.3390/app15126576
58
Turismo De PortugalI. P. (2020). 2019 activity report. Available online at: https://www.turismodeportugal.pt/SiteCollectionDocuments/gestao/Informacao-de-Gestao/relatorio-atividade-2019-com-anexos.pdf (Accessed February 20, 2026).
59
United Nations Development Programme - UNDP Brazil (2025). IDHM panel (1991-2021). Available online at: https://www.undp.org/pt/brazil/desenvolvimento-humano/painel-idhm (Accessed January 30, 2026).
60
United Nations (UN) (2025b). Sustainable Development: The 17 Goals (SDGs). New York, United States: United Nations, Department of Economic and Social Affairs. Available online at: https://sdgs.un.org/goals (Accessed March 6, 2026).
61
United Nations. (2025a). World Urbanization Prospects 2025: Summary of Results. Department of Economic and Social Affairs, Available online at: https://population.un.org/wup/ (Accessed April 20, 2026).
62
VasconcellosE. A. (2012). Urban Mobility and Citizenship, 2012. São Paulo: Senac Book.
63
VimalaT.PatilS. B.GaikwadG.DivekarR. (2026). “Sentiment-driven passenger satisfaction prediction in metro surveys using a hybrid model,” in International Journal of Engineering Research and Technology (IJERT), 15. 10.5281/zenodo.20665839
64
VoorbergW. H.BekkersV. J. J. M.TummersL. G. A. (2015). systematic review of co-creation and co-production: embarking on the social innovation journey. Public Manag. Rev.17 (9), 133–1357. 10.1080/14719037.2014.930505
65
XiongY.LiangJ.GuanC. (2025). Decoding public sentiment topics in google maps reviews on urban infrastructure development. Sci. Rep.15 (1), 21363. 10.1038/s41598-025-06451-6
Summary
Keywords
DeepSeek-V3, metro transportation system, public value, sentiment analysis, travel experience, urban mobility
Citation
Souza H, Costa Á and de Oliveira RR (2026) Operationalizing public value and the travel experience with artificial intelligence: a longitudinal analysis of user sentiment at subway stations in São Paulo and Porto. Front. Future Transp. 7:1924380. doi: 10.3389/ffutr.2026.1924380
Received
30 June 2026
Revised
27 July 2026
Accepted
03 August 2026
Published
18 September 2026
Volume
7 - 2026
Edited by
Daniela Anna Misul, Polytechnic University of Turin, Italy
Reviewed by
Yile Chen, Macau University of Science and Technology, Macao SAR, China
Idi Jahidi, Bina Nusantara University, Indonesia
Updates
Copyright
© 2026 Souza, Costa and de Oliveira.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Hadassa Souza, hadassa.abreu@aluno.ifsp.edu.br
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.