During this study, the four major databases of Web of Science (WoS), PubMed, Scopus, and Google Scholar were considered. The search found that WoS contains the most literature, and after screening and removing duplicates, the main source of literature is still WoS. WoS has higher quality literature and significantly better citation accuracy than other databases (Falagas et al., 2008; Wang and Waltman, 2016; Yuan et al., 2024). Therefore, this article chooses to extract data from the WoS database.

We obtained pertinent studies pertaining to AI and the mental health of college students from the WoS core dataset. In order to construct a comprehensive database for research on AI and mental health among college students, our retrieval process was centered on three key aspects. Firstly, to broadly obtain studies related to mental health, the WOS search expression was set as follows: TS = (“mental health” OR “mental disorder*” OR “mental wellness” OR “mental illness*” OR “depression” OR “stress” OR “anxiety” OR “bipolar disorder” OR “borderline personality disorder” OR “schizophrenia”). Secondly, to restrict the research object to college students, the WOS search expression was set as follows: TS = (“university student*” OR “college student*” OR “higher education” OR “undergrad student*” OR “master’s student*” OR “postsecondary education” OR “undergraduate*” OR “tertiary education” OR “postsecondary education” OR “doctoral student*” OR “Ph.D. student*”). Thirdly, to highlight the application of AI in college student mental health, the WOS search expression was set as follows: TS = (“machine learning” OR “artificial intelligence” OR “deep learn*” OR “neural network” OR “convolutional neural network” OR “recurrent neural networks” OR “natural language process*” OR “CNN” OR “LSTM” OR “RNN” OR “AI” OR “support vector machine” OR “logistic regression” OR “random forest” OR “naïve bayes” OR “decision tree”). The Boolean operator OR was used between keywords in all three searches, and Boolean operator AND was used between three topics. In addition, asterisks were used as wildcards to explain various variations of several keywords.

The screening methodologies were implemented as follows: Document type restrictions were applied to include “Article,” “Early Access,” “Book Chapter,” and “Conference Proceedings.” The time span was delimited from January 2003 to December 2023, and the language criterion was set to English. As of December 29, 2023, a total of 3,117 articles were retrieved. Furthermore, throughout the entire process of data screening and collection, two independent reviewers (J. Chen and D. Yuan) identified potential studies based on inclusion and exclusion criteria by scrutinizing article titles, abstracts, and full texts. Any disparities that emerged were deliberated upon and resolved through discussion, resulting in a consensus. All data for this study were extracted by December 31, 2023, and following the screening, a total of 1,722 articles were deemed eligible for inclusion. The data selection analysis process is delineated in Figure 1.

Bibliometrics, as a quantitative analysis of literature, serves as a methodological tool to delineate developmental trends within specific academic fields, emphasizing the attainment of quantifiable, replicable, and objective results. In this study, we conducted an analysis of annual publication counts and citation frequencies to elucidate the evolutionary patterns within the field over the past two decades. The growth rate, a pivotal metric portraying developmental trends (Guo et al., 2020), was calculated based on the increase in publications over time relative to the baseline year of 2003. It is noteworthy that while the growth rate signifies the rate of change in publications relative to 2003, it does not provide insight into the proportion of AI-related publications among all publications. Additionally, utilizing data from the (WoS), we examined citation data for the top 10 journals, research domains, countries, institutions, and authors with respect to publication quantity. This investigation aimed to provide insights into the distribution and predominant productivity of publications in the domain of AI and college student mental health.

Growth rate (%) = [(number of publications in the last year ÷ number of publications in the first year)^{1/(last year - first year)} - 1] × 100%.

We employed VOSviewer software, a widely recognized tool for bibliometric analysis, to visually examine literature. VOSviewer utilizes the interrelationships among academic entities such as authors, journals, and keywords to depict scientific research trends within a specific domain. Analysis units in VOSviewer encompass countries, authors, journals, keywords, and more, tailored to the analytical focus and database type. Nodes in the visualization represent analysis units, with node size indicating their significance in the network. Connections between nodes are depicted by lines, with thicker lines denoting stronger connections. Node colors denote distinct clusters, where nodes of the same color belong to the same cluster (Hernandez-Torrano et al., 2020). In this study, we initially analyzed collaborative networks in countries/territories and authors to elucidate patterns of scientific collaboration in AI and college student mental health. Subsequently, a co-citation analysis based on publication sources was conducted to reveal potential disciplinary compositions within this field. Finally, through visual examination of keywords, we explored research hotspots and developmental trends in the domain.

Figure 2 illustrates the annual evolution of publications and citation frequency associated with AI and college student mental health research from 2003 to 2023. The research trajectory in this field can be roughly divided into three stages: First Stage (2003–2007): During this initial period, the volume of studies focusing on AI and college student mental health was relatively modest. The growth rate from 2003 to 2007 stood at 7.46%, indicating a comparatively subdued level of attention and interest from researchers in this field. Second Stage (2008–2017): Noteworthy increases were observed in both publications and citation frequency during this phase, with a growth rate of 17.05%. This suggests that, with the development of AI, the application of AI in college student mental health gradually gained attention. Third Stage (2018–2023): Both the publications and citation frequency showed significant increases and publications during this timeframe constituted 75.45% of the total papers (1,337/1772), with a growth rate of 50.46% from 2018 to 2022. In 2022, the pinnacle was reached with the highest number of publications (410) and citation frequency (7362). While there was a slight downturn in both the annual publications and citations in 2023, they remained at an elevated level.

Keywords Are The author’s extraction and summarization of core terms In a research study. In this study, VOSviewer software Was used To calculate The frequency of keyword occurrence and Total link strength (TLS). TLS represents The total co-occurrence frequency of a keyword with other keywords; The larger The value, The greater The importance of The node In The network graph (Kuang et al., 2023). Table 4 Presents information about The frequency of keyword occurrence and TLS. In The field of AI and college student mental health, The Top five mental health issues Are depression, anxiety, stress, mental disorders, and major depression. Among them, depression Has The highest TLS of 872, indicating Its crucial role In college student mental health issues. The Top five AI technologies Are machine learning, AI, deep learning, random forest, and decision tree. Machine learning Has The highest TLS of 149, signifying Its prominence As The most important technology In this field. The Top five applications of AI In college student mental health Are risk factors, prediction, assessment, clustering, and digital health. Among them, risk factors have The highest TLS of 142, indicating that using AI technology To discover risk factors for college student mental health Is One of The most important applications In this field.

In this study, we employed VOSviewer software to visualize the global scientific collaboration network in the field of AI and college student mental health from 2003 to 2023, encompassing countries/territories and researchers. Figure 3 delineates the international collaboration network in this field, with lines connecting nodes indicating cooperative relationships among countries/territories. Thicker lines denote closer collaboration. This analysis includes only countries that have contributed a minimum of 6 articles to the dataset, resulting in 107 nodes and 535 connections. China and the United States, positioned as the largest nodes, hold central roles in the network graph, establishing robust collaborative ties with other countries or territories (such as Canada and South Korea) within the Light Blue cluster, underscoring their significant influence in this field.

Figure 4 depicts the collaboration network among researchers, where nodes represent researchers, the node size indicates the quantity of publications by researchers, and lines between nodes denote collaboration relationships. This analysis includes only authors with a minimum of 5 articles in the dataset, resulting in 55 nodes and 113 connections. The graph reveals the presence of multiple collaborative teams, with the Red cluster led by Mortier, Philippe and the Green cluster led by Wang, Qiang emerging as the two largest collaborative teams in this field. Researchers within these teams exhibit close connections, indicative of robust collaboration. Additionally, there are other collaborative teams with five or more members, including the Blue cluster led by Oh, Hans, the Purple cluster led by Yang, Tingzhong, and the Yellow cluster led by Tzourio, Christophe. Meanwhile, independent authors appear less frequently in the graph, and the thickness of the lines signifies the degree of association between researchers, contingent on the frequency of their co-appearance in the same articles. The position of researchers in the clusters reflects their frequent and interconnected collaboration with other researchers, aligning with collaboration patterns observed in other bibliometric studies where researchers with social importance and productivity tend to enhance the productivity of their co-authors (Ho et al., 2017).

The co-citation relationship between journals refers to two journals cited by the same third journal. The more frequently two journals are cited by other journals, the closer their co-citation relationship, and journals that are frequently co-cited are considered to have a strong theoretical and semantic foundation (Hernandez-Torrano et al., 2020). To comprehensively explore the disciplinary foundation of AI in the field of college student mental health, we conducted a co-citation analysis of journals in the dataset using VOSViewer. In this analysis, only journals cited at least 60 times were considered, resulting in 180 nodes and 15,268 connections, as show in Figure 5. The findings reveal that research on AI in the field of college student mental health encompasses multiple disciplines to some extent. Specifically, the knowledge in this field is constructed by integrating five interrelated disciplines. In the lower-right portion of the graph, the Red cluster predominantly brings together journals from the field of psychiatry, such as Lancet, Psychiatry Research, BMC Psychiatry, and others. In the lower-left portion, the Green cluster is mainly composed of journals from the fields of public health and behavioral science, including BMC Public Health, Journal of American College Health, Addictive Behaviors, and more. In the upper-right portion, the Yellow cluster signifies the contribution of computer science and interdisciplinary journals to this field, such as Computers in Human Behavior, Journal of Medical Internet Research, Nature, Science, and others. In the upper-middle portion, the Purple cluster integrates relevant journals from the field of psychology, such as Psychological Medicine. In the upper-left portion, the Blue cluster encompasses clinical psychology, child and adolescent psychiatry, and other psychiatry-related journals, including Journal of Clinical Psychology, Clinical Psychology Review, European Child & Adolescent Psychiatry, and more. In summary, research on AI in the field of college student mental health is interdisciplinary, primarily involving the integration of psychology, computer science, and public health.

To uncover the research hotspots in the field, we employed VOSviewer software for keyword co-occurrence analysis. Keyword co-occurrence refer to the frequency of two keywords appearing in the same article, and a network graph of keyword co-occurrence can delineate the research hotspots in a specific field. In this analysis, only keywords that appeared at least 10 times were considered, resulting in 82 nodes and 949 connections. Figure 6 illustrates the keyword co-occurrence graph and identifies five major research directions in the field of AI in college student mental health from 2003 to 2023.

The Purple cluster encompasses keywords like depression, anxiety, internet addiction, sleep, etc., representing the widespread attention of researchers to the impact of internet addiction on young people’s depression and anxiety. The Green cluster includes college students, depressive symptoms, physical activity, alcohol, etc., indicating that the mental health of college students is the second research hotspot in this field. The Yellow cluster comprises university students, stress, quality of life, COVID-19 pandemic, etc., suggesting that coping with the stress and quality of life of university students is the third research hotspot. The Blue cluster involves suicidal ideation, risk factors, mental distress, etc., representing factors posing a risk to the mental health of college students. The Red cluster involves keywords like mental health, machine learning, artificial intelligence, deep learning, etc., indicating that the application of AI technology in mental health is another widely discussed topic. Additionally, the five clusters are interconnected and closely related, signifying a close correlation among the five research directions.

To explore the temporal trends of research hotspots in this field from 2003 to 2023, we conducted a time-based visual analysis of keywords, as depicted in Figure 7. Deep blue represents early research hotspots, yellow represents recent research hotspots, and the node size signifies the level of research intensity, with the legend indicating the average year. It is evident that early research hotspots in this field primarily focused on aspects such as sleep, internet addiction, alcohol, risk factors, etc. Around 2020, the research hotspots shifted toward college students, depression, anxiety, and related topics. By around 2022, keywords like “machine learning,” “artificial intelligence,” and “deep learning” represented recent research hotspots.

This study employs bibliometric methods to explore the current status and trends of research on mental health and AI amomg college student by analyzing the development trends and distribution of publications in the WOS database, characteristics of research fields, scientific collaboration patterns, potential disciplinary compositions, and research hotspots and trends. In general, over the past two decades, the number of publications related to AI and college student mental health has steadily increased, reaching its peak in 2022, indicating a high research interest in this field. The journal with the most publications is the Journal of Affective Disorders, and Psychiatry Research has the highest average citation frequency. Psychiatry is the most significant research field. The IF and JCR partition are crucial indicators for assessing the quality of journals and the influence of papers. Papers published in high-IF journals or within the Q1/Q2 category are generally regarded as having higher academic standards and impact (Bornmann and Marx, 2016). In this study, a total of 721 journals published 1722 articles. The top 10 ranked journals published 452 articles, accounting for over a quarter of the total. Additionally, nine of these journals were positioned in either the Q1/Q2, and six had an IF exceeding 4. These data indicate that the data sources used are reliable and have a high academic level and attention. Additionally, China and the United States have made substantial contributions to the growth of publications in this field. Harvard University in the United States is the most influential institution, and the most prolific and impactful authors are Prof. Peltzer Karl and Kessler Ronald C. Keyword analysis indicates that the primary mental health issues faced by college students are depression, and the most widely used AI technology in college student mental health is machine learning, with the most applications being risk factors. Keyword co-occurrence analysis reveals that the research hotspots in this field mainly focus on the factors influencing college students’ mental health and the application of AI in college student mental health. Recently, high-frequency keywords such as “machine learning,” “artificial intelligence,” and “deep learning” represent the latest research trends and frontiers in this field.

Furthermore, collaboration in this field needs to be strengthened, including international and interdisciplinary collaboration. Figure 3 indicates that international collaboration in this domain is substantially shaped by language, geography, and cultural backgrounds. For example, the largest international collaboration network (Red cluster) comprises primarily 15 countries/territories from South America and Europe with similar cultural backgrounds, where English serves as the primary language. The second-largest international collaboration network (Green cluster) is comprised of 6 countries/territories from Asia, 3 from Europe, and 2 from Africa. The third-largest international collaboration network (Dark Blue cluster) encompasses 10 countries/territories from Asia, demonstrating geographic proximity. Collaboration mainly concentrates in Europe, the Americas, and Asia, with limited international collaboration among authors from developing countries, and even less contribution from authors in underdeveloped territories. Cross-national collaborative research has become a means to enhance research quality, increase resource utilization, and influence (Hsiehchen et al., 2015). The research gap between developing and developed countries and its impact on global health research trends cannot be ignored (Akinremi, 2011). Journal co-citation analysis shows that research in this field is, to some extent, interdisciplinary. Journals from various fields jointly constitute the disciplinary framework of AI and college student mental health research, primarily composed of psychology and computer science, supplemented by public health, behavioral science, and other health-related disciplines. Robust interdisciplinary partnerships integrating multilevel behavioral socioecological models with the molecular, cellular, and ultimately physiological bases of health and disease will contribute to breakthroughs in public health (Mabry et al., 2008). Therefore, we hope that in the future, developed countries can collaborate more with countries in underdeveloped territories in this field, and researchers from different disciplines can establish good collaborative relationships to produce higher-quality research results, making research in this field more universally applicable.

The co-occurrence analysis of keywords (Figure 6) reveals that the factors influencing college students’ mental health are one of the research hotspots in this field. Many factors contribute to the occurrence of mental health disorders in college students, including gender, genetics, personal factors, as well as family, school, and societal aspects (Beiter et al., 2015; Thompson et al., 2020; Seehuus et al., 2021). However, this study identified several interesting points.

Firstly, the keyword “internet addiction” in the Purple cluster caught our attention. Internet addiction profoundly affects the mental health of college students, as prolonged immersion in the virtual world can lead to anxiety, depression, and other psychological issues (Shen et al., 2020). A meta-analysis involving 33,650 college students showed a positive correlation between mobile phone addiction (MPA) and anxiety, depression, and sleep disturbances (Li et al., 2020), suggesting that students with MPA are more prone to high levels of anxiety and depression. Logistic analysis by Lin et al. (2019) confirmed a significant negative correlation between the degree of internet addiction and sleep quality, indicating that poor sleep quality increases the risk of psychological disorders in college students (Nyer et al., 2013). Additionally, the impact of internet addiction on college students’ mental health is evident in social isolation and self-identity. The popularity of social media applications has reduced face-to-face communication time in the real world, leading to social isolation in males who engage in online gaming and affecting self-identity in females who use the internet to avoid exclusion from online interactions. The social standards and values in the virtual world can influence college students’ self-identity, causing confusion and negative emotions (Tateno et al., 2019; Kato et al., 2020; Herruzo et al., 2023). Furthermore, research indicates a moderate negative correlation between exercise and internet addiction in young people, with the relationship’s strength unaffected by data collection time, country or region, or population type (Xiao et al., 2022). Exercise therapy has been shown to be effective in treating internet addiction in college students (Zhang et al., 2023). Therefore, incorporating regular physical activity into higher education policies or encouraging students to engage in sports can reduce the potential threat of internet addiction to mental health.

Secondly, the keyword “alcohol” in the Green cluster is another factor influencing the mental health of college students. Excessive alcohol consumption may lead to psychological problems such as anxiety and depression in college students (Schry and White, 2013). The impact of alcohol on college students’ mental health may primarily stem from its damage to the brain and nervous system. Long-term alcohol abuse can affect normal brain function, leading to cognitive decline, impacting academic performance, and increasing psychological burden (Meda et al., 2017; Zahr and Pfefferbaum, 2017). Additionally, researchers have identified the liver aldehyde dehydrogenase 2 (ALDH2) in the brain’s astrocytes as a primary mechanism for the direct regulation of alcohol-induced behavioral functional impairment. This provides strong evidence for the role of the brain in alcohol metabolism (Jin et al., 2021). Leclercq et al. (2020) transplanted the intestinal microbiota of alcohol use disorder patients into mice, resulting in significant social disorders in mice, demonstrating the association between gut microbiota imbalance and changes in brain alcohol metabolism. These changes, in turn, were related to a decrease in the synthesis of β-hydroxybutyrate (BHB) in the liver, hindering the neuroprotective effects of BHB. Moreover, alcohol abuse may disrupt sleep, affecting emotions and cognitive abilities, with negative effects on mental health (Lund et al., 2010). In summary, the harm caused by alcohol to college students’ mental health should not be overlooked. To maintain mental health, students should be vigilant about alcohol consumption, adopt a healthy lifestyle, seek social support, and, when necessary, seek professional psychological counseling. Through positive attitudes and scientific psychological adjustment methods, college students can better cope with the psychological challenges alcohol may pose.

Thirdly, the impact of the keyword “covid-19 pandemic” in the Yellow cluster on the mental health of college students is equally noteworthy. The COVID-19 pandemic and associated lockdowns have profoundly affected the mental health of college students, with the uncertainty and fear brought about by the pandemic leading to widespread anxiety and stress among students (Singh et al., 2020). A survey targeting college students revealed that, out of 7,143 respondents, 0.9% experienced severe anxiety, 2.7% experienced moderate anxiety, and 21.3% experienced mild anxiety (Cao et al., 2020). A broader survey reported that approximately one-fifth of Chinese college students exhibited symptoms of anxiety during the COVID-19 pandemic (Fu et al., 2021). Economic impacts, disruptions to daily life, and delays in academic activities emerged as major sources of psychological stress, posing a severe threat to the mental health of college students (Cao et al., 2020). Moreover, the COVID-19 pandemic resulted in social isolation, depriving college students of normal social activities. Online learning and restricted living conditions led to feelings of loneliness and social isolation among students, preventing them from enjoying traditional campus life. Prolonged social deprivation may contribute to the accumulation of emotional issues, negatively impacting the mental health of college students (Lim et al., 2022; Sun et al., 2023). Additionally, the pandemic heightened stress and burdens for college students, who may experience increased academic pressure and uncertainties about the future, leading to psychological issues such as anxiety and depression (Robledo-Martin et al., 2023) To safeguard the mental health of college students, it is crucial to adopt proactive psychological adjustment strategies. Recommendations include maintaining social connections, seeking social support, adopting a healthy lifestyle, and learning effective stress management and emotional regulation methods. The availability of mental health services in higher education institutions is also an essential resource, providing professional assistance and support to help students navigate through this challenging period (Oswalt et al., 2020; Kihumuro et al., 2022).

The co-occurrence analysis of keywords (Figure 6) reveals that the Red cluster signifies the application of AI technology in the field of mental health, representing another research hotspot. The keyword visualization analysis over time (Figure 7) identifies “machine learning,” “artificial intelligence,” and “deep learning” as the latest research trends and frontiers in this field. Through comprehensive data analysis and consensus among two article reviewers, we have identified four directions for AI in the domain of college students’ mental health: (1) prediction and diagnosis; (2) treatment; (3) prognosis; (4) ethical considerations. The following sections will discuss each category in detail.