Bibliometric Analysis of Research Progress in Tunnel Safety Based on CiteSpace

Summary

This study presents a bibliometric analysis of tunnel safety research, using Web of Science data and the CiteSpace tool to identify publication trends, collaboration, and research hotspots. The analysis reveals a focus on fire and traffic safety, shifting toward intelligent management and multi-hazard resilience, with fragmented collaboration.

Abstract

Due to the ever-increasing complexity of tunnel spaces and the diversification of associated risks, a comprehensive understanding of tunnel safety research is essential. This study employs CiteSpace to conduct a bibliometric analysis of publications on tunnel safety in the Web of Science Core Collection. The results show that the number of publications has increased steadily, with a significant surge after 2017. Central South University is the leading contributor, and Tunnelling and Underground Space Technology stands out as an important journal. The research is highly interdisciplinary, encompassing underground engineering, fire safety engineering and related disciplines, although collaboration networks among authors and institutions remain fragmented, reflecting limited synergy. Keyword analysis shows that research hotspots focus on tunnel fire safety and traffic accident analysis, with a transition from fundamental theory and risk assessment toward intelligent and system-oriented management. The field is shifting from traditional static analysis to dynamic operational safety analysis, multi-hazard coupled risk assessment, and system resilience evaluation. However, challenges remain in integrating digital technologies and transdisciplinary collaboration. Unlike traditional narrative reviews that often rely on fragmented or static assessments, this study provides a reproducible bibliometric framework for dynamically mapping the evolution of tunnel safety research. It offers methodologically structured insights for identifying emerging trends and guiding future interdisciplinary investigations.

Introduction

As a critical component of modern transportation and infrastructure networks, tunnels have increased continuously in scale and complexity worldwide. However, the unique and confined environment within tunnels presents significant safety challenges, including structural cracking¹, fire risks², traffic hazards³, operational safety concerns⁴, and issues related to whole-lifecycle safety monitoring⁵. Historical data consistently show that incidents within tunnels, especially fires, frequently lead to severe casualties and property damage⁶. These incidents not only damage the vehicles and the tunnel structure itself, but also reveal the inadequacies of existing safety response strategies⁷. Therefore, in-depth research on tunnel safety is not merely fundamental for ensuring the long-term stable operation of infrastructure, but also necessary for safeguarding public safety, optimizing emergency management systems⁸, promoting sustainable development⁹, and improving societal well-being.

Early research on tunnel safety can be traced back to the 1990s. The research topics mainly focused on fundamental issues such as the accumulation of flammable gases and personnel evacuation in tunnels. Subsequently, the field experienced significant progress, with numerous scholars conducting extensive and valuable research. The research scope gradually expanded to include issues such as tunnel accident mechanisms, fire protection system design, construction safety control, and structural design optimization. Since the beginning of the 21^st century, particularly over the past two decades, with the rapid advancement of digitalization and intelligent technologies, the research scope further evolved toward frontier areas such as structural stability monitoring^10,11, accident risk prediction, tunnel lighting systems, driver behavior, and driver information interaction systems. Existing research on tunnel safety can be broadly categorized into three main aspects.

First, some studies focus on the transmission mechanisms of fire disaster chains under multi-factor coupling conditions¹². These studies employed methods such as statistical accident analysis¹³, case-based retrospective studies, and probabilistic risk assessment (PRA) to investigate accident causation, risk evaluation¹⁴, fire evolution mechanisms, evacuation simulation¹⁵, and structural safety monitoring. Second, a number of studies focus on the enhancement of proactive prevention and control capabilities in dynamic environments. Many scholars have utilized computational fluid dynamics (CFD) simulations¹⁶, BIM-GIS integrated modeling, and multi-source sensor network technologies to analyze fire smoke propagation, visibility attenuation, dynamic evacuation path optimization, and health monitoring and early warning during tunnel operation¹⁷. These studies have further optimized tunnel ventilation and smoke exhaust systems¹⁸, improved intelligent lighting configurations¹⁹, and advanced the exploration of human behavior modeling and digital twin-driven closed-loop management. Third, certain studies primarily focus on the application of advanced technologies and the pursuit of sustainable tunnel development. Some scholars have employed digital and intelligent technologies such as Virtual Reality (VR)²⁰and Building Information Modeling (BIM)²¹for tunnel blasting²², underground void detection, and excavation support. These studies also explore the coordinated development of environmental, social, and economic factors during tunnel engineering processes, with a focus on sustainability²³. In addition, some scholars have applied bibliometric methods to review research on ground settlement in tunneling²⁴, conduct visual analyses of infrastructure inspection²⁵, and examine the relationship between tunnel lighting and low-carbon development²⁶.

Tunnel safety has evolved into a comprehensive and interdisciplinary research field^6,26. However, systematic and integrated studies are still lacking, and the evolution of the knowledge structure and identification of emerging trends in this field are still insufficiently clear, making it difficult to determine future research directions. Narrative reviews are valuable for interpreting mechanisms, comparing engineering practices, and synthesizing expert knowledge. However, they are often limited by the subjectivity of literature selection and by their difficulty in quantitatively tracing the temporal evolution of large research fields. Bibliometric analysis complements narrative review by providing transparent retrieval criteria, repeatable data-processing procedures, and quantitative indicators of collaboration, centrality, clustering, and burst evolution. In this study, CiteSpace was used not as a substitute for expert interpretation but as a visualization and quantitative-mapping tool to identify the intellectual structure and emerging themes of tunnel safety research. The scope of this study was limited to tunnel safety. This scope included fire safety, smoke control, evacuation, traffic safety, driver behavior, tunnel lighting, ventilation, environmental control, monitoring, emergency management, risk assessment, and system resilience during tunnel operation. Studies primarily concerned with tunnel excavation safety, shield-tunnel construction prediction, blasting, ground settlement, construction support, and lining-crack diagnosis were excluded unless they explicitly addressed operational safety outcomes. These exclusions were applied because construction-stage geotechnical risk and operation-stage safety management differ substantially in mechanisms, data sources, evaluation indicators, and engineering interventions^27,28.

To systematically clarify the characteristics and trends of research in the field of tunnel safety, this study conducts a bibliometric analysis using the CiteSpace visualization tool based on data from the Web of Science Core Collection (WoSCC), excluding Chinese-language publications and various informal research materials. The analysis covers publication output characteristics, major journals, collaboration networks among core authors and institutions, the distribution and clustering of research keywords over time, and emerging frontier directions²⁸. In contrast to conventional review approaches, this study establishes a quantitative, reproducible, and visualization-based bibliometric protocol. It integrates multi-dimensional metrics, such as keyword burst detection, timeline clustering, and co-occurrence centrality analysis. Using these metrics, the study constructs a comprehensive and dynamic knowledge framework that systematically clarifies the field’s intellectual structure and thematic evolution. This study has two primary analytical objectives. The first objective is to construct a comprehensive and dynamic knowledge framework using the CiteSpace visualization tool, thereby overcoming the fragmented and static nature of traditional review assessments. The second is to investigate whether tunnel safety research has undergone identifiable phase transitions, specifically a shift from conventional fundamental theories and static risk analysis toward an intelligent, systematic, and multi‑hazard coupled dynamic safety management system.

Protocol

This study conducted a systematic bibliometric analysis to identify and visualize research trends in the field of tunnel safety from 1992 to 2026.

Literature search and data collection
Access the Web of Science Core Collection (WoSCC) database. The search strategy was performed using the following topic-specific query: TS = [(“tunnel” and “tunnel safety”)]. After the query was entered, the “Search” button was clicked to execute the literature search. The results were filtered by selecting “Articles”, “Review Articles”, and “Early Access Publications” as the document types. A total of 15861 articles and review articles published between 1992 and 2026 were collected. This study focuses on core areas such as tunnel fire safety, road safety, traffic safety, and personnel evacuation, aiming to review the research progress and hotspots in overall tunnel safety. Therefore, publications related to excavation safety of tunnels, safety prediction of shield tunnels, and safety assessment of tunnel lining cracks were excluded. Following this screening process, 428 articles were retained for further analysis. In the “Record Content” section, “Full Record and Cited References” was selected, and the filtered dataset was exported as “Plain Text” files. Each record was verified to contain complete citation and co-authorship information. The detailed screening process is presented in a flow diagram (Figure 1).

Data preprocessing
The plain text dataset exported from WoSCC was imported into a spreadsheet using the Data | From Text/CSV function. Tab and space characters were selected as delimiters to separate the text into columns. The resulting spreadsheet was used to filter, sort, and summarize the records for subsequent analyses. Records were aggregated by publication year to calculate annual and cumulative publication counts. The Journal (SO) column was filtered to retain only journals with three or more articles. Institutions were sorted by frequency and centrality to identify the top 10 institutions, and keywords were sorted by frequency and centrality to identify the top 20 keywords. Keyword clusters were sorted by size, and the six largest clusters were retained. Detailed criteria and procedures for each operation were described in the corresponding analytical subsections.

For the analysis of the number evolution of published literature, the Publication Year (PY) column was filtered in the spreadsheet software, and the values were converted to numeric format. The records were then aggregated by publication year to calculate both annual and cumulative publication counts. A publication trend graph was generated by constructing a bar chart for annual output and overlaying a line chart for the cumulative total (Figure 2). For the analysis of distribution patterns of publication journals, the Journal (SO) column was isolated, and the number of records for each journal was calculated. Publication frequency was calculated for each journal between 1992 and 2026 by grouping records according to journal title and sorting the results in descending order. Journals with three or more articles were retained, resulting in 28 journals, and the findings were presented in a summarized table (Table 1).

Authors’ collaboration analysis
The WoSCC plain text dataset was imported and configured in CiteSpace using the Data | Import | Web of Science command, and the input and output directories were specified as appropriate. Following an in-depth review of the retrieved literature, publications from 1992 to 2004 were found to account for a relatively small proportion of the dataset. More importantly, the research content and themes of these publications were dispersed and showed weak relevance to tunnel safety. Incorporating these papers into the bibliometric analysis would have introduced excessive redundancy and obscured the identification of research hotspots and evolving trends. In addition, the literature retrieved in March 2026 could not fully represent the annual publication output for 2026. Consequently, the CiteSpace analysis was restricted to literature published between 2005 and 2025. The time span was set from 2005 to 2025 based on publication relevance, using annual time slicing with a time slice of 1 year.

CiteSpace was configured with author as the node type. The selection criterion was set to the g-index (k = 15). Term sources were set to Title, Abstract, Author Keywords (DE), and Keywords Plus (ID). Under the Pruning menu, both the Pathfinder and Pruning the merged network options were selected to simplify the network structure. The analysis was executed by clicking Go to produce the author collaboration network visualization. Font sizes and color schemes were adjusted to improve readability (Figure 3).

Institutional collaboration and top 10 institutions analysis
The node type was set to institution in CiteSpace. The selection criterion was set to the g-index (k = 25). Term sources were set to Title, Abstract, Author Keywords (DE), and Keywords Plus (ID). Under the Pruning menu, the Pathfinder and Pruning the merged network options were selected to reduce network complexity.

The analysis was run by pressing Go to generate the institutional collaboration map. Typography and color settings were adjusted to improve interpretability (Figure 4). The institutional collaboration data were saved using Output | Network Summary as HTML, CSV, RIS | Save as CSV. The CSV file was then opened in Microsoft Excel. The “Freq”, “Centrality”, and “Label” columns were combined, and a descending sort was applied using Sort & Filter | Descending. The top 10 institutions were extracted and assembled into a final table (Table 2).

Keyword co-occurrence and top 20 keywords analysis
In CiteSpace, the node type was set to keyword. The g-index was set to k = 17. Term sources were set to Title, Abstract, Author Keywords (DE), and Keywords Plus (ID). Under the Pruning menu, the Pathfinder and Pruning the merged network options were selected to trim the network. The analysis was initiated by clicking Go to obtain the keyword co-occurrence map. Font and color attributes were modified to improve visual readability (Figure 5). The keyword co-occurrence dataset was exported through Output | Network Summary as HTML, CSV, RIS | Save as CSV. The CSV file was opened in Microsoft Excel. The “Freq”, “Centrality”, and “Label” columns were combined, and a descending sort was performed using Sort & Filter | Descending. The top 20 keywords were identified and organized into a final table (Table 3).

Keyword cluster and top 6 clusters analysis
The node category was set to keyword in CiteSpace. The g-index was set to k = 17. Term sources were set to Title, Abstract, Author Keywords (DE), and Keywords Plus (ID). Under the Pruning menu, the Pathfinder and Pruning the merged network options were selected to refine the network. Keyword clustering analysis was performed by clicking Go. Cluster labels were extracted using the log-likelihood ratio (LLR) algorithm. Only clusters with a silhouette score greater than 0.7 were retained. The clustering map was generated using Cluster Label Optimization | K: Keywords, and the display settings were adjusted for optimal readability (Figure 6). Cluster details were extracted via Clusters | View Cluster Content and organized into a Microsoft Excel worksheet. The data table was opened, and the “Size” column was selected and sorted using Sort & Filter > Descending. The results were compiled into the final keyword clustering table (Table 4).

Keyword timeline cluster analysis
The node type was kept as a keyword in CiteSpace. The g-index was set to k = 17. Term sources were set to Title, Abstract, Author Keywords (DE), and Keywords Plus (ID). Under the Pruning menu, the Pathfinder and Pruning the merged network options were selected to prune the network. Keyword clustering analysis was executed by clicking Go. Cluster labels were extracted using the same LLR algorithm and silhouette score threshold (> 0.7). Timeline View was selected to generate the keyword timeline clustering map. Font and color settings were adjusted to enhance readability (Figure 7).

Keyword bursting analysis
The node type was kept as a keyword in CiteSpace. The g-index was set to k = 17. Term sources were set to Title, Abstract, Author Keywords (DE), and Keywords Plus (ID). Under the Pruning menu, the Pathfinder and Pruning the merged network options were selected for network pruning. Keyword analysis was executed by clicking Go. The Burstness option was selected, the parameter γ was set to 0.5, and the minimum burst duration was set to 2 years to perform the keyword burst detection analysis. The data were refreshed to generate the list of the top 25 keywords with the strongest citation bursts (Figure 8).

Results

Publication output and growth trends
The initial search retrieved 15861 articles and reviews related to tunnel safety published between 1992 and 2026 from the Web of Science Core Collection (as of 26 March 2026). After manually excluding publications unrelated to the research focus (e.g., studies on tunnel excavation safety, tunnel safety prediction, and tunnel lining crack safety assessment), 428 articles and reviews focusing on tunnel safety were retained for bibliometric analysis. This refined dataset ensured a targeted examination of the literature in the field of tunnel safety.

The annual number of publications and its evolution trend can indicate the prominence and activity within a research field²⁹. Microsoft Excel was used to compile statistics on the annual and cumulative numbers of publications on tunnel safety from 1992 to 2026 (Figure 2). The results show that, since 1992, research on tunnel safety has been published internationally. Overall, the research on tunnel safety can be categorized into three different phases: the budding phase (1992–2004), the slow development phase (2005–2016), and the rapid development phase (2017–2026). The proportions of the literature published in these three phases are 5.607%, 19.626%, and 74.766%, respectively. From 1992 to 2004, the published literature was extremely scarce, with a total of 24 publications in 13 years. In this phase, the literature briefly discussed the theory of road safety in tunnels and fire safety, indicating that scholars were just beginning to pay attention to tunnel safety. During 2005 to 2016, the number of publications increased gradually, with a total of 84 articles in 12 years. In this phase, the topics of the articles gradually expanded, such as tunnel safety, risk analysis, tunnel fire, and tunnel energy consumption. During 2017 to 2026, there were 320 articles in 10 years, and the research scope expanded significantly, encompassing simulation, computational fluid dynamics (CFD), risk assessment, human evacuation behavior, intelligent ventilation, and traffic accidents. Especially in recent years, with the rapid development of computers, the Internet of Things (IoT), and robotics, researchers have increasingly applied IoT and robot technology in disaster response, real-time monitoring, and data acquisition, aiming to achieve intelligent safety assessment and risk control.

Journal publication analysis
To characterize the distribution of these publications, we counted the journal sources of the 428 articles from 1992 to 2026. It was found that there were 147 journals publishing literature related to tunnel safety. As shown in Table 1, journals with three or more publications were presented. The results showed that 28 journals were identified. “Tunnelling and Underground Space Technology” led with 76 articles, far more than any other journal, followed by “Fire Safety Journal” with 33 articles. Thus, it can be seen that the field of underground engineering and geotechnics has conducted more in-depth research on tunnel safety, followed by the fields of fire protection engineering and traffic safety management, reflecting the multidisciplinary and cross-disciplinary nature of the research.

Analysis of authors’ collaboration
Investigating authors’ publication collaborations helps identify the major researchers and teams in the field. Figure 3 shows the author collaboration network map generated by CiteSpace. The size of the nodes denotes the number of articles published by each author, and the connecting lines denote collaboration relationships. The color of the connecting lines represents the publication year of articles; darker colors signify later publications. Each node denotes an author, and the connecting lines signify direct collaborations formed in the articles. The author collaboration network map showed 291 nodes and 310 connecting lines, and a density of 0.0073. This indicated that cooperation and communication among the authors were not close, and further efforts are needed to enhance this aspect. In the author collaboration network map, five relatively large collaborative research teams were identified. The most active collaborative research teams in the research field of tunnel safety mainly included Du Zhigang, Yang Yongzheng, Han Lei, and He Shiming from China. Their research provided a comprehensive analysis of visual guiding systems for enhancing traffic safety in freeway tunnels and investigated accidents at highway tunnel exits^30,31,32,33. Haukur Ingason and Li Yingzhen from RISE Research Institutes of Sweden, together with Bjelland Henrik from Norway, focused on the redirection of smoke flow in inclined tunnel fires^34,35. Another prominent group mainly consisted of Konstantinos Kirytopoulos, Huang Helai, and Konstantinos Kazaras, who revealed several gaps in drivers’ knowledge regarding tunnel safety and equipment, noting that drivers often adopt inappropriate habits and behaviors while driving through tunnels^4,36. At Lund University in Sweden, Enrico Ronchi, Hakan Frantzich, and Karl Fridolf conducted extensive research on the evacuation safety measures in large infrastructure tunnel projects^37,38. A group from China, mainly including Zhu Hehua, Huang Xinyan, Zhang Yuxin, and Shen Yi, conducted in-depth research on the smart real-time evaluation of tunnel fire risk and evacuation safety via computer vision³⁹.

Moreover, according to the statistical results, Du Zhigang and Haukur Ingason were the top two authors in terms of publication output, each with 23 articles. In light of Price’s law, the minimum number of papers published by the core authors of tunnel safety was 3.59⁴⁰. Rounding to an integer, authors who had published more than 4 articles were regarded as the core authors in this field. There were only 19 authors who had published more than 4 articles. This demonstrated that a core group of authors has begun to emerge, although its scale remains limited.

Overall, dispersed research perspectives and the limited number of core authors indicate that this research field still requires further integration and more focused research⁴⁰. Meanwhile, in the future, researchers should give full play to their own core advantages and strengthen collaboration and exchange with other authors from different countries and institutions, so as to further promote improvement in tunnel safety research.

Analysis of institutional collaboration
Investigating the current status and collaboration patterns of research institutions helps clarify future directions for institutional collaboration and in-depth research in this field. As shown in Table 2, Central South University had the most papers with a frequency of 23, followed by Wuhan University of Technology (22). Next were Chang'an University, Southwest Jiaotong University, and Tongji University, with 20, 19, and 18 articles, respectively. Centrality quantitatively signifies the influence of the institution. Central South University is the most influential institution with a centrality of 0.09. A secondary tier of influential institutions, each with a centrality of 0.08, included Huazhong University of Science & Technology, Chinese Academy of Sciences, Southwest Jiaotong University, Shanghai Jiao Tong University, and Qingdao University of Technology. Next were the Hong Kong Polytechnic University, State University System of Florida, Nanyang Technological University, and Lanzhou University of Technology, with a centrality of 0.07.

As shown in Figure 4, the institution cooperation network map showed 217 nodes and 208 connecting lines, and a density of 0.0089. This indicated that cooperation and communication among the institutions were not close, and further efforts are needed to strengthen this aspect. Central South University was identified as the most influential research institution, with Huazhong University of Science & Technology, the Chinese Academy of Sciences, Southwest Jiaotong University, Shanghai Jiao Tong University, and Qingdao University of Technology forming a secondary influential cohort. Notably, some highly productive institutions did not exhibit correspondingly high centrality. This indicates that while they are active in research, their role as collaborative hubs is less pronounced. Therefore, fostering deeper collaboration is crucial for exchanging theoretical insights and advancing the field.

Keyword co-occurrence analysis
Keywords can indicate the main topics of a research field. The analysis of keyword co-occurrence and frequency can reveal research hotspots^28,40. Keyword co-occurrence in the related literature was investigated. As illustrated in Figure 5, each node denotes a keyword, the size of nodes denotes the occurrence frequency of the keywords; the connecting lines between two nodes signify connections between the two keywords. The color of the nodes represents the occurrence time of the keywords, with red denoting the earliest occurrence and purple denoting the latest. The keyword co-occurrence analysis map showed 293 nodes and 585 connecting lines, and a density of 0.0137. The relatively low network density in the published literature indicated weaker correlations among keywords. This indicated an expanding scope of tunnel safety research. Yet the current stage was still accumulative and characterized by independent multidisciplinary exploration. Further in-depth investigation is needed, which is especially true for the synergistic evolution mechanisms of multiple factors in complex disaster chains.

The top 20 keywords with the highest frequency and centrality are listed in Table 3. It is important to note that a keyword centrality greater than 0.1 indicates that it is significant, highlighting its influence within the published literature network, often due to high connectivity or continual citations by other keywords. The results showed that the combined frequency of “road tunnel” and “road tunnels” was 74. They were followed by “traffic safety” and “tunnel fire,” each with a frequency of 37. In addition, the frequencies of “design,” “model,” and “traffic accident” were also relatively high, at 34, 30, and 29, respectively. The centrality values of “traffic safety” and “fire” were the highest, with a centrality of 0.32. Other keywords with high centrality include “accidents”, “road tunnel” and “tunnel safety”, indicating that the field places primary emphasis on fire-related risks and traffic safety. In addition, “flow”, “behavior”, and “environment” also rank highly in centrality, suggesting that safety assessment, human behavior, and environmental control play key roles in linking different research areas. Moreover, keywords such as “critical velocity”, “smoke temperature”, “tunnel entrance”, and “CO yield”, each with a centrality of 0.09, suggested that recent research increasingly focused on topics such as driving behavior simulation in tunnels and smoke control and ventilation in tunnel fire scenarios.

Keyword cluster analysis
The keyword clustering module in CiteSpace enables the identification of high-frequency terms within the scholarly literature and is widely employed for keyword co-occurrence cluster analysis. This functionality operates on the Log-Likelihood Ratio (LLR) algorithm. Using these calculations, the algorithm assigns cluster labels that are characterized by minimal redundancy and strong alignment with real-world research contexts. Consequently, this approach facilitates the elucidation of thematic architectures, emerging research foci, and evolutionary trajectories within a given domain. The 11 cluster labels presented in the diagram are: heat release rate, road tunnels, traffic safety, risk analysis, energy savings, virtual reality, fire suppression, high-speed train, thermal safety, risk assessment, and safety factor (Figure 6). In the co-cited network map, the modularity Q value of the clustering map was 0.782, and the silhouette S value was 0.927, indicating that the clustering results are significant and reasonable. Keywords within the same cluster are strongly correlated, while those between different clusters show good independence.

For this study, 6 cluster labels were selected for subsequent analysis (Table 4). Among them, “#0 heat release rate” was the largest cluster in the knowledge map, followed by “#1 road tunnels” (32), “#2 traffic safety” (29), “#3 risk analysis” (27), indicating that research related to fire dynamics and tunnel safety constitutes the core thematic focus in this field, while traffic safety and risk analysis also represent important subtopics with substantial research activity.

Keyword timeline cluster analysis
Figure 7 shows the evolution of keywords over time in the research on tunnel safety. The 11 labels from #0 to #10 on the right side indicate the research themes that are the clusters of the keywords in published literature. The keywords were sorted by their year of appearance and classified into the corresponding clusters. The connecting lines between keywords indicate co-occurrence of the two keywords in the same literature. The size of nodes indicates the occurrence frequency of the keywords in this cluster.

As shown in Figure 7, tunnel safety has been the main focus of the related research since 2005. From 2005 to 2010, the main keywords included “traffic safety”, “road tunnel”, “tunnel fire”, “design”, “longitudinal ventilation”, indicating that research focused on tunnel safety, fire prevention, and ventilation design^41,42. During 2011 to 2015, with the continuous development of theoretical research, studies on traffic accidents and human behavior in fires became a new research hotspot, mainly concentrating on evacuation behavior and psychological mechanisms⁴³. “Energy savings,” “behavior,” and “exit choice” gradually became research focuses after 2015, reflecting increasing attention to evacuation behavior, human responses in fires, and energy-efficient ventilation strategies in road tunnel fire safety research⁴⁴. After 2016, the types of keywords and research themes increased significantly, indicating that tunnel safety research had entered a new stage. During this period, keywords including “driving safety”, “tunnel lighting”, “impact”, “performance”, “speed” and “temperature” became research hotspots. This suggested that the research focus expanded beyond basic tunnel safety mechanisms to driving safety, tunnel environmental monitoring and operational risk control⁴⁵.

Overall, tunnel safety research focused on three main subjects: the tunnel structure itself, the surrounding environment, and drivers. In terms of tunnel operation, research primarily addresses fire safety and operational risk assessment. Within the field of highway tunnel safety evaluation, studies have progressively advanced toward intelligent real-time assessment techniques, quantitative risk assessment methods, and analytical approaches based on artificial intelligence, such as flow network algorithms. Research on drivers mainly aims to ensure driving safety and has increasingly focused on the effects of spatial and visual conditions within tunnels on driving behavior in recent years. With the development of information and biomedical technologies, driver performance under different conditions is assessed using devices such as eye-tracking systems. In addition, the scope of research has expanded to safety issues under complex tunnel conditions, including long tunnels and underwater tunnels as representative challenging environments.

Keyword bursting analysis
The sudden emergence of keywords and their variant forms may indicate a shift in research hotspots within a given field. Thus, using the keyword burst analysis module in CiteSpace, this study presents the top 25 keywords exhibiting the strongest citation bursts in the published literature (Figure 8). It should be noted that the “intensity” in Figure 8 represents burst intensity, with a higher intensity indicating greater impact. The dark blue band in the figure denotes the overall time span of the keyword, while the red band indicates the duration during which the keyword becomes a research hotspot. The results indicate that, excluding “tunnel safety” (with a strength of 2.56), the top five keywords with the highest burst strength are “tunnel fire” (4.00), “heat release rate” (3.98), “system” (2.83), “impact” (2.47), and “road tunnel” (2.43), indicating their prominence in tunnel safety research. The keywords with the longest duration of bursts are “tunnel safety” (9 years), “heat release rate” (9 years), “fire” (8 years), “risk analysis” (7 years), as well as “road tunnel”, “exit choice”, and “systems”, all of which maintained a burst duration of five years. Overall, the three main stages in the research field of tunnel safety are summarized as follows:

From 2005 to 2010, the research focused on tunnel fire safety and basic risk theory. Key hot keywords during this period included “tunnel safety”, “risk analysis”, “tunnel fire” and “fire”. The release of The Handbook of Tunnel Fire Safety in 2005 drew significant attention to tunnel safety and fire safety. Notably, the keyword “tunnel fire” began to burst prominently in 2010, indicating that the academic community attached great importance to tunnel fire safety and risk management.

During 2011 to 2016, the research hotspots shifted to fire and smoke control as well as ventilation strategies in road tunnels. Representative keywords included “critical ventilation velocity”, “road tunnel”, “exit choice”, “longitudinal ventilation” and “energy savings”. The shift of research focus during this period was driven, on the one hand, by the implementation of the European Union (EU) Construction Products Regulation (CPR, Regulation (EU) No 305/2011)^46,47, which applies to all construction works including tunnels. On the other hand, it was affected by several typical tunnel fire accidents, such as the 2015 Skatestraum Tunnel fire in Norway and the 1996 and 2015 Channel Tunnel fires^48,49. These events exposed critical vulnerabilities in existing safety systems and catalyzed a paradigm shift in policy and regulatory frameworks.

From 2017 to 2025, research in the field of tunnel safety focused on intelligent safety management and systematic risk prevention and control throughout the entire tunnel lifecycle. Key efforts included real-time monitoring of critical environmental parameters during tunnel operation, strengthening safety measures in accident-prone areas such as tunnel entrances, analyzing the comprehensive impact of tunnel safety on surrounding environments and traffic operations, and establishing scientific assessment systems. Hot keywords during this period included “systems”, “tunnel entrance”, “temperature”, “safety”, “speed”, “impact” and “construction”. Since 2023, the prominence of the keyword “construction” suggests that it may become a future research hotspot and trend in tunnel safety.

DATA AVAILABILITY
The Web of Science Core Collection bibliographic records retained after screening and used for the bibliometric analysis are provided as Supplemental File 1. The CiteSpace software parameter settings used to generate the visualizations are described in the Protocol section. These materials support transparency and reproducibility of the reported bibliometric results.

Figure 1. Flowchart of the screening process. Please click here to view a larger version of this figure.

Figure 2. Annual and cumulative numbers of published literature during 1992 to 2026. From 1992 to 2026, both the annual and cumulative numbers of related publications showed an upward trend, and the literature output entered a phase of rapid growth after 2020. Please click here to view a larger version of this figure.

Figure 3. Cooperation network map of authors of published literature. The size of the nodes denotes the number of articles published by each author, and the connecting lines denote collaboration relationships. The color of the connecting lines represents the publication year of articles, darker colors signify later publications. Please click here to view a larger version of this figure.

Figure 4. Cooperation network map of institutions of published literature. This figure visualizes the institutional cooperation network of published literature, showing major research institutions and their collaborative relationships in the field. Please click here to view a larger version of this figure.

Figure 5. Co-occurrence network map of keywords in published literature. Each node denotes a keyword, the size of nodes denotes the occurrence frequency of the keywords, the connecting lines between two nodes signify connections between the two keywords. The color of the nodes represents the occurrence time of the keywords, with red denoting the earliest occurrence and purple denoting the latest. Please click here to view a larger version of this figure.

Figure 6. Keyword clustering map of published literature. Polygons are color-coded to differentiate distinct thematic domains, with red representing the largest cluster and blue indicating smaller clusters. Please click here to view a larger version of this figure.

Figure 7. Timeline diagram of keywords in published literature. This timeline map shows keyword evolution, with node colors indicating publication year (red for recent years, purple for earlier years) and node size representing frequency. Please click here to view a larger version of this figure.

Figure 8. Burst map of keywords in published literature. The horizontal bar chart shows the top 25 keywords that experienced marked increases in frequency, reflecting the evolving research trends and shifting priorities in tunnel safety research from 2005 to 2025. Please click here to view a larger version of this figure.

Journal name	Publication number of literature	Rank
Tunnelling and Underground Space Technology	76	1
Fire Safety Journal	33	2
Traffic Injury Prevention	17	3
Accident Analysis and Prevention	16	4
Applied Sciences-Basel	14	5
Safety Science	14	5
Sustainability	12	7
Fire Technology	9	8
Transportation Research Record	8	9
Fire-Switzerland	7	10
International Journal of Hydrogen Energy	7	10
Journal of Transportation Safety & Security	7	10
Transportation Research Part F-Traffic Psychology and Behaviour	6	13
Buildings	5	14
International Journal of Environmental Research and Public Health	5	14
Proceedings of the Institution of Civil Engineers-Civil Engineering	5	14
Reliability Engineering & System Safety	5	14
Underground Space	5	14
Advances in Mechanical Engineering	4	19
Process Safety and Environmental Protection	4	19
Energy	3	21
Gallerie E Grandi Opere Sotterranee	3	21
IEEE Access	3	21
International Journal of Thermal Sciences	3	21
Journal of Transportation Engineering Part A-Systems	3	21
Proceedings of the Institution of Civil Engineers-Transport	3	21
Proceedings of the Institution of Mechanical Engineers Part F-Journal of Rail and Rapid Transit	3	21
Thermal Science	3	21

Table 1: Statistics on the number of literature published in journals. This table summarizes the names and publication counts of the top 28 journals ranked by the number of published articles.

No	Institution	Frequency	Institution	Centrality
1	Central South University	23	Central South University	0.09
2	Wuhan University of Technology	22	Huazhong University of Science & Technology	0.08
3	Chang'an University	20	Chinese Academy of Sciences	0.08
4	Southwest Jiaotong University	19	Southwest Jiaotong University	0.08
5	Tongji University	18	Shanghai Jiao Tong University	0.08
6	RISE Research Institutes of Sweden	17	Qingdao University of Technology	0.08
7	Chongqing Jiaotong University	16	Hong Kong Polytechnic University	0.07
8	Hong Kong Polytechnic University	12	State University System of Florida	0.07
9	Beijing University of Technology	10	Nanyang Technological University	0.07
10	National Technical University of Athens	10	Lanzhou University of Technology	0.07

Table 2: Statistics on the frequency and centrality of top 10 institutions. This table summarizes the top 10 research institutions ranked by frequency and by centrality, along with their respective frequency or centrality.

No	Keyword	Frequency	Keyword	Centrality
1	Road Tunnel	43	Traffic Safety	0.32
2	Traffic Safety	37	Fire	0.32
3	Tunnel Fire	37	Accidents	0.28
4	Design	34	Road Tunnels	0.28
5	Road Tunnels	31	Tunnel Safety	0.23
6	Model	30	Flow	0.21
7	Traffic Accidents	29	Behavior	0.21
8	Safety	25	Environment	0.18
9	Behavior	24	Longitudinal Ventilation	0.17
10	Longitudinal Ventilation	22	Model	0.15
11	Tunnel Safety	22	Tunnel Fire	0.14
12	Flow	21	Risk Analysis	0.13
13	Simulation	18	Road Tunnel	0.13
14	Impact	18	System	0.12
15	System	17	Safety Evaluation	0.11
16	Performance	17	Fire Safety	0.1
17	Fire Safety	16	Critical Velocity	0.09
18	Critical Velocity	14	Smoke Temperature	0.09
19	Risk Assessment	12	Tunnel Entrance	0.09
20	Accidents	11	CO Yield	0.09

Table 3: Statistics on the frequency and centrality of top 20 keywords. This table summarizes the top 20 keywords ranked by frequency and by centrality, along with their respective frequency or centrality.

Cluster ID	Size	Sihouette	Mean (Year)	Top Terms (LLR, P-Level)
Heat Release Rate	40	0.883	2015	Heat Release Rate (13.24, 0.001); Tunnel Fire (11.76, 0.001); Longitudinal Ventilation (8.97, 0.005); Critical Velocity (8.97, 0.005); Backlayering (8.81, 0.005)
Road Tunnels	32	0.97	2015	Road Tunnels (11.45, 0.001); Safety Management (7.85, 0.01); Tunnel Fire (7.4, 0.01); Driving Safety (5.67, 0.05); Case Study (4.34, 0.05)
Traffic Safety	29	0.934	2019	Traffic Safety (33.32, 1.0E-4); Tunnel Lighting (9.59, 0.005); Tunnel Fire (7.81, 0.01); Lighting Attenuation (7.66, 0.01); Driving Behavior (7.66, 0.01)
Risk Analysis	27	0.934	2011	Risk Analysis (15.62, 1.0E-4); Tunnel Safety (14.28, 0.001); Disaster Warning (8.72, 0.005); Emergency Evacuation (8.72, 0.005); Adjacent Construction (8.72, 0.005)
Energy Savings	24	0.992	2017	Energy Savings (22.1, 1.0E-4); Lighting (10.98, 0.001); Sustainability (10.98, 0.001); Dialux (10.98, 0.001); Transmittance (5.47, 0.05)
Virtual Reality	24	0.982	2015	Virtual Reality (14.78, 0.001); Human Behaviour in Fire (6.19, 0.05); Utility Tunnel (6.19, 0.05); Visual Features (4.9, 0.05); Behavioural Training (4.9, 0.05)

Table 4: Top 6 clusters of Keywords in the Field of Tunnel Safety. This table presents the top six keyword clusters in the tunnel safety research field, along with their size, silhouette, mean year, and top terms.

Supplemental File 1. Web of Science Core Collection records used for the bibliometric analysis. This plain-text file contains the 428 records retained after screening from the Web of Science Core Collection search. The records were exported with full record information and cited references and were retrieved on March 26, 2026. This file was used as the source dataset for the publication-trend, journal-distribution, collaboration-network, keyword co-occurrence, keyword clustering, timeline-cluster, and keyword burst-detection analyses described in the protocol. Please click here to download this file.

Discussion

This bibliometric analysis provides a structured overview of research trends in tunnel safety. By integrating multi-dimensional metrics including keyword burst detection, timeline clustering, and co-occurrence centrality analysis, this study constructs a dynamic knowledge framework. This framework clarifies the field’s intellectual structure and thematic evolution. Unlike traditional narrative reviews that often rely on fragmented or static assessments, this quantitative and visualization-based protocol offers a reproducible model for mapping scientific progress in tunnel safety research. Its transparent and repeatable workflow also makes it easily adaptable to other engineering safety topics.

Several methodological steps strongly affect the validity of the bibliometric results. The first critical step is database retrieval, because changes in the search query, database scope, document type, and retrieval date directly affect the final dataset. The second critical step is manual screening, especially the distinction between tunnel safety and construction-stage tunnel engineering. The third critical step is keyword normalization, because unmerged variants such as “road tunnel” and “road tunnels” can artificially split important topics. The fourth critical step is parameter selection in CiteSpace, including time slicing, node-selection criteria, term sources, and pruning methods. These settings influence network density, cluster stability, and the visibility of emerging topics. Therefore, retrieval decisions, screening decisions, keyword-merging rules, and software parameters should be reported transparently.

Some common problems may occur in bibliometric analysis. If WoSCC records are incomplete, the dataset should be re-exported using “Full Record and Cited References,” and the presence of abstracts, keywords, author affiliations, and cited references should be checked. If duplicate records are detected, they should be removed by comparing accession numbers, titles, digital object identifiers, and first-author information. If the collaboration or keyword network is too dense, consistent pruning should be applied to improve readability. If the network becomes too sparse, the node-selection threshold should be adjusted and the parameter change should be reported. If clustering results are unstable, modularity Q, silhouette S, cluster labels, and representative terms should be compared across parameter settings before drawing conclusions.

CiteSpace parameter choices can influence the resulting networks and interpretations. A broader time span or a larger node-selection threshold usually increases the number of nodes and links but may reduce readability. Stronger pruning improves visual clarity but may remove weak links that represent emerging interdisciplinary topics. Term-source selection also affects the results. Using only author keywords may emphasize author-defined topics, whereas adding abstracts and Keywords Plus can reveal broader conceptual associations. Burst-detection settings affect which terms are identified as emerging hotspots. Therefore, the main conclusions should be interpreted together with the reported parameters, and sensitivity checks should be performed when the network structure changes substantially.

The usability of this bibliometric framework lies in its transparent and repeatable workflow. Researchers can reproduce the analysis by using the documented search query, retrieval date, eligibility criteria, keyword-normalization rules, software parameters, and exported data tables. The framework can also be adapted to other engineering safety topics by modifying the search query and redefining inclusion and exclusion criteria. Admittedly, due to some randomness in the algorithm employed by CiteSpace, the visualization results may exhibit minor variations across multiple runs. Meanwhile, the method cannot replace experimental fire testing, traffic simulation, structural safety calculation, accident reconstruction, or predictive safety modeling. Its main function is to map the knowledge structure of a research field and to guide more focused qualitative or quantitative investigations.

This framework not only captures macro-level shifts in research focus, such as the transition from single-hazard risk to multi-hazard resilience, but also identifies specific bridging points where emerging themes remain weakly coupled with core safety clusters. For instance, the weak connection between the “energy savings” cluster and the core safety-related clusters should be interpreted cautiously. This pattern may reflect a genuine research gap because many studies on tunnel energy saving focus on lighting control, ventilation efficiency, and operational cost reduction without explicitly linking these strategies to safety-performance indicators. However, it may also be partly influenced by the clustering algorithm and keyword-normalization strategy, because energy-related studies may use terms such as “lighting,” “ventilation,” “visual comfort,” or “low carbon” rather than explicit safety keywords. Therefore, the result should be interpreted as evidence of a potentially underconnected research direction rather than definitive proof of disciplinary separation. Future studies should more explicitly examine trade-offs and synergies between energy-efficient operation and safety requirements, such as smoke-control effectiveness, visibility, evacuation time, and driver visual adaptation.

Based on the bibliometric results and existing research findings, current hotspots in the field mainly focus on tunnel fire dynamics and longitudinal ventilation, traffic safety and driver behavior, intelligent risk assessment and real-time control, and safety enhancement under complex tunnel operating scenarios. Numerical simulation tools, particularly CFD, have become the primary research methods in this domain. Scholars simulate airflow characteristics, temperature distribution, and the diffusion patterns of hazardous gases under fire scenarios.

They also examine the evolution of oxygen and carbon dioxide concentration fields under different ventilation strategies. These efforts help investigate fire control measures and evacuation processes in tunnels^{16,50,51,52,53}. Due to the enclosed nature of tunnels, fires can easily create high-temperature environments and smoke accumulation. This leads to reduced visibility and severe consequences. Since 2005, “tunnel safety” and “risk analysis” have emerged as initial hotspots. Subsequently, starting from 2010, “tunnel fires”, “critical ventilation velocity” and “longitudinal ventilation” have remained persistent research focuses. Studies have gradually shifted from ventilation system design to fire dynamic mechanisms and smoke control strategies^16,54. They have also moved toward accident prevention and disaster mechanism analysis⁵⁵. Quantitative risk assessment of fire evolution has also become an important research direction. The keyword timeline diagram shows that “driver” and “behavior” appeared before 2015 and were early research focuses. After 2015, “driving safety” and “tunnel lighting safety” appeared more frequently. This indicates a shift in research perspective from infrastructure-focused studies to complex system safety analyses centered on human behavior^56,57,58,59. Some studies emphasize drivers' operational behavior, particularly visual adaptation, cognitive load, and behavioral changes at critical locations such as tunnel entrances and exits. Their aim is to improve safety through optimized lighting design and traffic signage⁶⁰. Others focus on fire simulation, using technologies such as virtual reality to model evacuation processes. They analyze individual route choice, panic behavior, and group interactions. This provides experimental support for emergency evacuation system design.

Research on energy efficiency and sustainable operation in tunnels has gradually emerged as a relatively distinct cluster. However, its integration with mainstream safety research remains limited. The burst analysis reveals that "energy savings" became a prominent keyword from 2015 to 2019, with a burst strength of 2.41 (Figure 8). This direction aims to optimize ventilation systems, lighting control strategies, and overall energy consumption during tunnel operation. These aspects are closely associated with both reducing operational costs and achieving environmental sustainability. However, as shown in Figure 6, the “energy savings” cluster shows limited connections with the “Virtual Reality” and “High-speed train” clusters, and even weaker links with safety-related clusters. This means that studies on energy conservation tend to focus more on independent energy efficiency optimization. In the field of tunnel safety, the core focus of “energy savings” lies in energy-efficient tunnel lighting and overall energy efficiency optimization. For example, during driving, technologies such as lighting simulation or transmittance optimization are used to improve the tunnel lighting environment, thereby enhancing tunnel safety. If energy saving is directly linked to tunnel safety, the current research is still limited. There seem to be few studies that clearly combine energy-saving strategies with reliable safety performance standards. Bridging this gap means shifting from treating energy savings as an isolated operational goal to a collaborative optimization objective within a safety framework.

With the rapid advancement of digital technologies, emerging hotspots increasingly concentrate on tunnel risk assessment and real-time control. Researchers employ artificial intelligence, big data, and the Internet of Things to analyze factors such as smoke⁶¹, temperature, and ventilation conditions. This enables real-time identification of fire scenarios and accurate prediction of fire development. Intelligent algorithms are further used to support scientific tunnel safety design. They promote a transition from static evaluation to intelligent decision-making^62,63,64,65. Moreover, with the growing number of long, large-section, and underwater tunnels, coupled multi-hazard risks and safety issues under extreme conditions have drawn increasing research attention^66,67.

According to the bibliometric results and the temporal evolution of keywords, tunnel safety research has evolved from addressing single issues to dealing with multi-factor coupled systems. Early studies mainly focused on theoretical analysis and engineering experience. The emphasis was on tunnel structures, ventilation, and fire prevention and control. As theoretical frameworks advanced, numerical simulation techniques were gradually introduced to enable quantitative risk assessment. In recent years, research attention has increasingly shifted toward data-driven approaches and artificial intelligence. The focus is now on dynamic coupling mechanisms in complex tunnel projects and full-process emergency management. Overall, the field has moved from traditional experience-based analysis toward data-driven and intelligent decision-making. Both the depth and breadth of research have grown significantly. Future research in tunnel safety is expected to focus on the following aspects.

First, in terms of theoretical development, substantial progress has been made in tunnel fire safety research. This has played an important role in engineering practice. However, the current perspective is still largely confined to single-hazard fire scenarios. Existing studies primarily address smoke ventilation efficiency, human evacuation behavior, and structural stability under high temperatures. Less attention has been given to structural damage and collapse risk under seismic loading, flooding at tunnel portals induced by extreme rainfall, performance degradation of underground structures, and risk assessment under multi-hazard spatiotemporal coupling^68,69. This limitation restricts the robustness of the theoretical framework under complex and extreme environments. It also hinders the development of resilience-oriented tunnel engineering across the full life cycle. Future research should adopt a more comprehensive, multi-dimensional perspective. The goal is to establish an integrated theoretical framework and robust evaluation models. Greater emphasis should be placed on disaster mechanisms, multi-factor coupling, and system-level safety. Key directions include structural damage and emergency strategies under strong earthquakes, data-driven simulation of cascading hazard events, assessment of flood impacts on tunnel performance, and optimization of rapid drainage systems. Reliability-based methods should also be introduced. These methods can reveal the performance stability and service life of tunnels under multi-hazard interactions. This will enable a shift from single-hazard prevention to multi-hazard resilience enhancement.

Second, in the context of digital and intelligent development, deeper integration of artificial intelligence, big data, and digital twin technologies into tunnel safety systems is required. Current studies have explored applications in tunnel operation monitoring⁷⁰, driving simulation, and construction management. For example, numerical methods have been used to detect tunnel fires in real time and reconstruct accident scenarios. Driver physiological and cognitive load models have been developed using eye-tracking and EEG data. Machine learning techniques have been applied for intelligent construction monitoring. However, several limitations remain. At the data level, there is a lack of unified standards for data acquisition and integration, which hinders effective data fusion and collaborative analysis. In terms of hazard response, most studies still rely on reactive or preventive maintenance strategies. The capability for proactive risk prediction remains insufficient. The integration between intelligent technologies and core safety requirements is still limited. Future research should focus on real-time early warning of traffic collision risks, dynamic structural health monitoring of tunnels⁷¹, and integrated intelligent diagnosis and collaborative decision-making across the full life cycle. This will support a transition from passive response to proactive prevention. Furthermore, the integration of energy-efficient operation with intelligent safety management systems should receive greater attention. An example is adaptive ventilation systems that simultaneously optimize smoke control and energy consumption.

Third, regarding safety challenges and resilience assessment in complex tunnel engineering, the increasing development of long tunnels, large cross-section tunnels, and underwater tunnels has introduced new difficulties. These challenges arise from their complex structures⁷², confined environments, and unique hazard evolution processes. Current research on complex tunnels mainly focuses on fire safety and evacuation, structural fire resistance, and ventilation and lighting systems⁷³. However, studies on emergency response in complex tunnel accidents, development of fire detection systems, and integrated risk assessment remain insufficient. These assessments should cover natural hazards, human-induced damage, and operational failures. Moreover, although current tunnel design standards incorporate risk assessment concepts and address major hazard types, risk management is still largely based on generalized safety measures. Differentiated emergency response strategies and unified evaluation standards for different hazard types have not yet been established. Therefore, future research should prioritize improvements in ventilation control strategies, evacuation efficiency, structural damage evolution and monitoring, emergency management systems, and evaluation standards for these special tunnel types.

Fourth, strengthening interdisciplinary collaboration and international cooperation should be further prioritized. Tunnel safety research currently spans multiple disciplines, including civil engineering, fire science, transportation engineering, and computer science. However, the depth of cross-disciplinary integration and system-level convergence remains limited. Future efforts should focus on building more closely connected trans-regional collaboration networks. This will enable effective sharing of multi-source knowledge and methodological approaches. Greater attention should also be given to incorporating advanced experiences from different countries and regions in tunnel safety regulations, design concepts, and emergency management practices. Such developments will help enhance coordination across research systems, improve consistency of standards, and strengthen overall resilience. This will improve the systematic and forward-looking nature of research in this field.

However, some limitations of this study warrant careful consideration. One notable concern is the exclusive dependence on a single literature database. Because this study drew solely from the Web of Science Core Collection, relevant work published in non-English outlets or indexed in alternative platforms such as Scopus may have been overlooked. This selection constraint risks underrepresenting contributions from regions where English is not the primary academic language. In addition, several methodological steps strongly affect the validity of the bibliometric results. The first critical step is database retrieval, as variations in the search query, database scope, document types, and retrieval date directly shape the final dataset. The second is manual screening, particularly the need to distinguish between tunnel safety and construction-stage tunnel engineering. The third is keyword normalization, because unmerged variants such as “road tunnel” and “road tunnels” can artificially split important topics. The fourth is parameter selection in CiteSpace. CiteSpace parameter choices can influence the resulting networks and interpretations. Changes in time slicing or selection criteria directly affect network density and cluster stability. For instance, a lower threshold may generate many small, unstable clusters, whereas a very high threshold may suppress emerging topics.

A further challenge lies in the nature of citation-based evaluation itself. Citation counts can be shaped by factors unrelated to research quality. These include the prestige of the publishing journal, patterns of self-citation, and the geographic concentration of scholarly networks. Such influences may overstate the visibility of established centers and understate equally rigorous work from less prominent institutions or emerging research communities. In light of these caveats, the quantitative trends reported here should be interpreted as one layer of evidence. They are best used together with qualitative expert review and broader literature synthesis to form a more balanced understanding of the field.

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
CiteSpace	Chaomei Chen	https://citespace.podia.com/	Bibliometric visualization software used to generate collaboration networks, keyword co-occurrence maps, cluster maps, timeline views, and burst-detection outputs.
Microsoft Excel	Microsoft Corporation	https://www.microsoft.com/excel	Spreadsheet software used for data cleaning, sorting, frequency calculations, and table preparation.
Web of Science Core Collection	Clarivate	https://webofscience.clarivate.cn/wos/woscc/basic-search	Bibliographic database used to retrieve the records included in the bibliometric analysis.