A CORRELATION ANALYSIS METHOD BETWEEN STUDENTS’ DIGITAL LITERACY AND MENTAL HEALTH

1. Introduction

With the rapid development of information technology, digital literacy has become one of the basic abilities that individuals must possess in modern society. Digital literacy not only involves basic computer operation skills but also includes comprehensive literacy in information retrieval, data processing, network security awareness, etc. Specifically, digital literacy covers key areas such as computer operation skills, information retrieval and management, data processing and analysis, and cybersecurity awareness. Master efficient information retrieval skills to quickly find what you need in massive amounts of information, thereby greatly improving learning and work efficiency. Data processing and analysis capabilities are not limited to simple data collection but also include the ability to discover problems and propose solutions through data analysis, which is particularly important in the era of data-driven decision-making. In addition, good network security awareness can protect personal privacy and data security and avoid unnecessary losses and risks. A high level of digital literacy can significantly improve an individual’s learning efficiency, workability, and social adaptability, and is a key factor in promoting all-around personal development. On the other hand, mental health is an important factor affecting individual quality of life and social adaptation. Mental health is not only about an individual’s emotional regulation and mental state but also about social functioning and life satisfaction. People with good mental health can effectively manage and regulate their emotions, maintain mental balance and stability, have positive self-perception and an optimistic attitude, and cope with the pressure and challenges in life. At the same time, people with good mental health perform well in social interactions, can establish and maintain healthy interpersonal relationships, improve social communication skills and teamwork skills, and thus find balance and satisfaction in work and life. Due to the fast-paced nature of modern living and the growing pressure to succeed, mental health issues among college students have become more prevalent and are receiving significant societal attention. Contemporary culture presents college students with a multitude of obstacles, including academic stress, fierce job market competitiveness, and social ambiguity. These variables could potentially harm their mental well-being. Hence, it is crucial to prioritize and enhance the psychological well-being of university students for the sake of their individual growth and societal cohesion. By exploring the correlation between digital literacy and mental health among college students, this study aims to provide a reference for educators and mental health professionals to help them design more effective education and intervention measures. Improving students’ digital literacy will not only enhance their academic and professional abilities but may also have a positive impact on their mental health, thereby promoting their overall development.

In recent years, more and more studies have begun to focus on digital literacy and mental health and the relationship between the two. De León et al. aimed to develop a digital literacy assessment tool as a diagnostic tool for pre-service teachers’ digital literacy and digital teaching ability. By collecting data from quantitative pilot studies and qualitative expert panel reviews, the results showed that the tool had good internal consistency and content validity.¹ Avdeeva and Tarasova examined several international assessment approaches and put up a conceptual framework that considers digital literacy as a multifaceted underlying structure. They also outlined the steps involved in creating a digital literacy assessment tool using principled assessment design and evidence-centered design methodologies.² Bearman et al. introduced an organizational framework to investigate the intricate correlation between digital technologies and assessment design. The framework examines the three primary functions of digital technology in assessment: as a tool for enhancing assessment, as a method for cultivating and validating digital literacy, and as a way of cultivating and validating distinct human capabilities.³ Hernández-Marín et al. analyzed 34 studies that used quantitative tools to assess students’ and teachers’ performance in media, information, and digital literacy (MIDL) in formal education. Studies have found that college students are the most widely studied group, especially in Spain and Mexico. The studies used the Teachers’ Common Digital Competence Framework, the ALFAMED questionnaire, and the IL-HUMASS questionnaire as core analysis tools, which showed good performance in assessing reliability and validity.⁴ Seboka et al. investigated the mental health literacy levels and information acquisition behaviors of Ethiopian college students in resource-limited environments. The results showed that mental health literacy was significantly associated with factors such as gender, digital health literacy, information acquisition behaviors, and family history of mental illness.⁵ Some studies have shown that high levels of digital literacy may have a positive impact on mental health, such as reducing the incidence of anxiety and depression and enhancing self-efficacy and social support. However, some studies have pointed out that excessive reliance on digital technology may lead to information overload, Internet addiction, and other problems, which may hurt mental health. Therefore, the relationship between digital literacy and mental health still needs to be further explored and verified.

This study aims to explore the correlation between digital literacy and the mental health of college students. By using the methods of Twin Support Vector Machine (TWSVM) classification and correlation analysis, the potential relationship between the two is revealed by systematically collecting and analyzing the digital literacy data and mental health data of college students. Specifically, this study will collect digital literacy data through questionnaire surveys and classify them using TWSVM. Then, the electroencephalogram (EEG) signals of the same group will be classified, and finally, the correlation between the two will be analyzed through chi-square validation. This study is of great significance in terms of theory and practical application. Theoretically, this study helps to deepen the understanding of the relationship between digital literacy and mental health and enriches the theoretical basis of related research fields. In terms of practical application, the results of this study can provide a reference for college educators and mental health professionals to help them design more effective education and intervention measures to improve the digital literacy level and mental health of college students. By improving digital literacy, students’ academic and professional abilities can be enhanced, and their mental health may also have a positive impact, thereby promoting their all-round development. The main contributions of this study are

(1)	Revealing the potential relationship between digital literacy and mental health by systematically collecting and analyzing relevant data.
(2)	Deepening the understanding of the relationship between digital literacy and mental health, enriching the theoretical basis of related research fields.
(3)	Providing practical references for college educators and mental health professionals to design more effective education and intervention measures, thereby promoting the all-round development of college students.

2. Literature Review

2.1. Defining and assessing digital literacy

The investigation on digital literacy commenced in 2006 and has subsequently progressed over the ensuing years. The main keywords of the initial research results are “information literacy” and “digital literacy education”. Researchers have examined the implied meaning of digital literacy, the structure of digital literacy, and the implementation of digital literacy instruction. After 2012, with the development of digital media, the study of “media literacy” was introduced. After 2018, “digital life”, “digital citizenship” and “digital economy” have become new research keywords, further enriching the content of “digital literacy” research. Recently, there has been an emergence of evaluative studies on digital literacy, with a specific focus on “digital competence” and “digital skills”. Currently, the notion of digital literacy encompasses not just a fundamental understanding of computers and the Internet, but also the competencies and proficiencies necessary to utilize digital devices and the Internet for purposes of study, production, communication, and enjoyment. Therefore, the use of digital technology is considered to be “a basic learning tool” and a comprehensive application ability. According to the research conclusion of Law et al., “digital literacy” encompasses the proficiency to effectively and securely utilize digital technology to acquire, manage, comprehend, integrate, communicate, evaluate, and generate information, with the ultimate goal of fostering employment, labor, and entrepreneurship. It encompasses a range of skills, including proficiency in using computers, the ability to effectively find and evaluate information, the capacity to critically analyze media, and the skill of understanding and utilizing information. Based on the aforementioned research findings regarding the concept of digital literacy, this study posits that “digital literacy” encompasses three specific talents that together provide a full literacy suited for the digital era. The first aspect is the utilization of data tools, which refers to the capacity to obtain, handle, recognize, combine, and convey digital resources. The second is the ability to use innovative thinking to create digital content and creatively solve learning and work problems. The third is safety ethics awareness, comprehensive professional literacy of lifelong learning and cultivation, etc.

At present, the European Framework for the Development and Understanding of Digital Competencies led by the European Union, and the Global Digital Literacy Skills Reference Framework revised by UNESCO are universal digital literacy frameworks with high international recognition. Among them, the EU-led Comprehensive Digital Competence Framework is a guiding framework for the development of digital education in Europe. Its latest version is DigComp 2.2 released in 2022. This version outlines five domains of proficiency, encompassing information and data literacy, effective communication and collaboration, digital content generation, digital security, and problem-solving. DigComp is a comprehensive framework that consists of five main categories, which are then broken into 21 more specific competencies, resulting in a highly thorough and complete structure.

2.2. Definition and assessment of mental health

Mental health refers to the good state of an individual in terms of emotions, behaviors, and social functions, and the ability to adapt to the environment, cope with stress, and experience happiness in life. The mental health of college students is specifically manifested in the ability to maintain a positive and stable attitude in learning, life, interpersonal relationships, and self-identity, and to have good emotional management and coping abilities, to effectively cope with academic pressure, interpersonal challenges and future planning and other challenges, and promote personal all-round development. The existing mental health assessment methods are mainly shown in Table 1.

2.3. Research on the relationship between digital literacy and mental health

Digital literacy¹⁸ has a significant impact on mental health. A high level of digital literacy can help individuals obtain information and solve problems more effectively, thereby reducing anxiety and stress caused by information asymmetry. The study by Sundell et al. explored the relationship between health literacy and digital health information search behavior among highly educated people and found that digital health information search behavior is closely related to higher health literacy. Digital literacy is increasingly important in the field of education increase.¹⁹ In their study, Audrin et al. systematically reviewed digital literacy in education through text mining technology and pointed out that digital literacy not only promotes academic development but also has a positive impact on students’ mental health.²⁰

The differences and innovations of this study compared with existing studies are as follows: First, existing studies mostly focus on traditional assessment methods, while this study will innovatively introduce machine learning methods to provide more accurate and dynamic assessment results by analyzing students’ online behavior and mental health data. Second, existing studies mostly focus on the general population or specific occupational groups, while this study pays special attention to the digital literacy and mental health of college students. College students face the dual challenges of academic pressure and career choice, and their digital literacy and mental health are more complex and worthy of in-depth discussion. Finally, this study combines the interdisciplinary methods of education, psychology, and computer science to explore the relationship between digital literacy and mental health from multiple dimensions, which can not only reveal a more comprehensive relationship mechanism but also put forward suggestions and countermeasures with practical guidance.

3. Research Methods

3.1. Data collection

This study mainly collects digital literacy assessment data and EEG data of the subjects in this group. For the quality literacy assessment data, the experimental data is mainly obtained by questionnaire. The design of the questionnaire should focus on the different dimensions of digital literacy, including information literacy, media literacy, technology literacy, and network literacy. A simple example of the questionnaire survey is shown in Table 2.

The collection of EEG data for the mental health experiment is as follows: the subjects are 100 college students aged 18 to 25 who took the questionnaire survey mentioned above. The experiment was conducted in a quiet laboratory environment, using the Neuroscan SynAmps RT EEG system, a 32-lead electrode cap, a sampling frequency of 1000Hz, and a 0.1–50Hz bandpass filter. The experiment consists of two stages: first, 5min of baseline EEG data is collected from the subjects in a sitting state; then, the subjects are asked to complete a series of digital tasks including information search, media evaluation, and online learning, each task lasting 10min, while EEG data is recorded. Thirty minutes of EEG data were collected from each subject.

3.2. Data preprocessing

In the preprocessing of the questionnaire survey data, the completeness of the questionnaire filling was first checked, and the questionnaires that were not completed or had obvious abnormal answers were deleted. Then the missing values were processed, and the mean filling method or interpolation method was used for a small number of missing values. Next, the scores of different dimensions were standardized (such as converting 1–5 scores to 0–1 intervals) for comprehensive analysis. Qualitative data (such as gender) were encoded into quantitative data (such as gender coded as 0 for male and 1 for female) for statistical analysis. Finally, 92 valid questionnaires were obtained after the above preprocessing steps. This study invited five experts in related fields to classify the digital literacy of these 92 subjects into two categories: high literacy and low literacy.

The EEG data of 92 subjects whose questionnaires were valid were selected. The preprocessing process of EEG data first includes data filtering, using a bandpass filter (such as 0.5–40Hz) to filter out low-frequency drift and high-frequency noise in the EEG signal. Then, eye movement and electromyographic artifacts were removed by electrooculogram (EOG) and independent component analysis (ICA) methods. Then, the continuous EEG signal is segmented by task, and each segment corresponds to a specific task state. Subsequently, time domain and frequency domain features, such as power spectral density (PSD), are extracted to characterize EEG activity under different task states. Finally, the feature data are standardized (such as Z-score standardization) to facilitate subsequent machine learning and statistical analysis. For each subject, they were also asked to fill out a mental health questionnaire (MHQ) during the experiment. The results of the questionnaire were used to determine whether the subject was mentally healthy, thereby determining the mental health label of each subject.

3.3. Classification

For the classification of questionnaire data and EEG data, this study uses SVM as the basic model. The objective function of traditional SVM is a single objective function. Although the standard SVM is suitable for the classification of small sample data, small sample data is prone to overfitting. In addition, when there are many data samples, the SVM with a single objective function will have the disadvantages of slow running speed, memory usage, and low generalization ability. To address this issue, Jimenez-Castaño et al. introduced the concept of TWSVM.²¹ The objective of this model is to identify two hyperplanes in n-dimensional space that are not parallel. Each hyperplane should be positioned as closely as feasible to the samples belonging to the same class, while simultaneously being positioned as far away as possible from the plane containing the samples of the other class. The objective function formula is as follows :

TWSVM can be summarized as the following two quadratic programming problems :

and where w is the normal vector of the separating hyperplane, b is the bias term, ${\xi }_i$ is the normal vector of the separating hyperplane; $p+q$ is the total number of samples and $d_i\ge 0$, $i=1,2$ is the penalty factor. Analyzing these two functions, we find that the objective function is the same as the traditional SVM, and is measured by the sum of Euclidean distances. Therefore, for the model, the purpose of minimizing each objective function is to ensure that the solved hyperplane is as close to the corresponding sample as possible. At the same time, the data of another type of sample is placed in the corresponding constraint condition, so that it can be ensured to be as far away from the hyperplane where the samples of the other type are located as possible. Assuming that the samples of the two types are equal, the training time of TWSVM is one-fourth that of SVM.

4. Experimental Results

4.1. Experimental description

The experiments in this study are mainly divided into three parts, namely digital literacy assessment, mental health identification, and the correlation analysis between the two. To compare the performance of the models selected in the two experiments of digital literacy assessment and mental health identification, several comparison models are introduced: SVM,²² random forest (RF),²³ logistic regression (LR),²⁴ decision tree (DT),²⁵ K-nearest neighbors (KNNs).²⁶ The parameter settings of each model are shown in Table 3. Since the essence of the two experiments is classification, the evaluation indicators used in this study are Accuracy, Precision, Recall, and F1-score.

The hardware used in this study was a desktop computer equipped with an Intel Core i7 processor and 16GB RAM. Data was collected using a 64-channel EEG system from Brain Products with a sampling rate of 1000Hz. The data acquisition and processing software was BrainVision Recorder and BrainVision Analyzer. Data analysis and model training were performed in Python programming language (version 3.8), using open-source libraries such as NumPy, SciPy, and scikit-learn for data processing and machine learning model training.

4.2. Digital literacy assessment results

According to the data collection in Sec. 3.1, 92 samples were obtained. Sixty samples were used as training samples and 32 samples were used as test samples. The number of categories is 2, representing high literacy and low literacy, respectively. High literacy refers to having good information technology capabilities and network security awareness, being able to retrieve and process information efficiently, and being proficient in using various digital tools. Low literacy refers to weak information technology capabilities, a lack of basic information retrieval and processing capabilities, and weak network security awareness. The digital literacy assessment results obtained on each model are shown in Table 4.

To more vividly compare the performance of each model on different indicators, Fig. 1 shows a comparison chart of the experimental results based on Table 4.

Based on the data shown in Table 4, it can be seen that TWSVM performs best in terms of accuracy (94.85%), precision (92.97%), and F1-score (93.61%), far exceeding other models such as LR, SVM, and DT. This reflects the superiority of TWSVM in handling high and low digital literacy classification tasks. It can effectively divide complex feature spaces and improve classification accuracy and generalization capabilities. In contrast, KNN performs the worst, which is lower than other models on all evaluation indicators, possibly due to its dependence on local neighboring samples, which leads to unstable classification results when the boundaries are unclear.

The comparison chart shown in Fig. 1 shows the performance differences of each model in accuracy, precision, recall, and F1-score. The figure clearly shows the significant advantages of TWSVM on all metrics, the balance between stability and performance of LR and SVM, and the obvious disadvantage of KNN. These results not only highlight the importance of selecting appropriate models for digital literacy assessment but also provide directions for further optimization of classification algorithms and feature engineering to improve the accuracy and practicality of assessment.

4.3. Mental health identification results

Based on the EEG data collected in Sec. 3.1, the window sliding method is used to take every 1min of EEG data as a sample. There are 92 subjects, and each subject randomly selects 10 sample data, so a total of 920 samples are obtained. Among them, 600 samples are used as training samples, and 320 samples are used as test samples. The number of categories is 2, representing healthy and unhealthy, respectively. The experimental results obtained by each model are shown in Table 5.

To more vividly compare the performance of each model on different indicators, Fig. 2 shows a comparison chart of the experimental results based on Table 5.

Analysis of the experimental data in Table 5 shows that TWSVM performs best in the mental health recognition task, with an accuracy rate of 95.89%, a precision rate of 94.91%, a recall rate of 92.35%, and an F1-score of 93.61%. The superiority of TWSVM stems from its ability to efficiently handle complex nonlinear features. By maximizing the application of interval classification and kernel techniques, it can effectively distinguish between mental health and unhealthy states, improving the generalization ability and stability of the model. Second, LR also performs well in accuracy (94.43%), precision (93.61%), recall (92.39%), and F1-score (93.00%), but its ability to handle nonlinear features is slight. It is inferior to TWSVM, mainly due to the slight gap in the classification accuracy of complex data sets. The worst model is KNN. Although it performs well in accuracy and precision, its recall and F1-score are significantly lower than other models, which is mainly affected by its processing of high-dimensional data and complex features restrictions.

The performance differences of each model on different indicators can be further observed through the drawn line chart. TWSVM maintains significant advantages in all evaluation metrics, especially outstanding performance in recall and F1-score, which shows its effectiveness and reliability in complex data analysis. In contrast, LR performs well when dealing with linearly separable problems but is slightly insufficient in nonlinear situations, while KNN’s limitations in high-dimensional data and complex features cause it to lag behind other models in overall performance. Therefore, choosing a TWSVM model suitable for task characteristics can maximize the accuracy and accuracy of mental health identification, providing important guidance for further research and practice.

4.4. Correlation analysis

Based on the experiments in Secs. 4.2 and 4.3, the digital literacy assessment and mental health identification results of each sample were obtained. To analyze the relationship between digital literacy and mental health, this paper uses the chi-square verification analysis method. The p-value of the chi-square test is 0.0025278. The correlation analysis results are shown in Fig. 3.

Based on the analysis results of the above code, we obtained the p-value of the chi-square test to evaluate the relationship between numerical literacy (high versus low) and mental health (healthy versus unhealthy). Based on the experimental results, the following conclusions can be drawn:

(1)

The p-value of the chi-square test is used to measure the degree of deviation between the observed data and the expected data, thereby assessing the correlation between digital literacy and mental health. If the p-value is less than the set significance level (set to 0.05 in this paper), then the null hypothesis can be rejected, that is, there is a significant correlation between digital literacy and mental health.

(2)

The stacked bar chart clearly shows the distribution of high digital literacy and low digital literacy groups in mentally healthy and mentally unhealthy states. Charts allow you to visually compare the differences between the two groups and observe the distribution proportion of mental health status within each group. Specifically, we can see that the number of people with a relatively healthy mental health status in the high digital literacy group is higher than in the low digital literacy group.

5. Conclusions

This paper studies the correlation between college students’ digital literacy and mental health and proposes a correlation analysis method based on TWSVM classification and chi-square test. Use TWSVM to classify the collected digital literacy data and obtain digital literacy assessment results. At the same time, EEG signals from the same group are collected, feature extraction based on PSD is performed, and the TWSVM classification model is trained to identify the mental health status of each student. On this basis, the chi-square test method was used to conduct a correlation analysis between the digital literacy assessment results and the mental health identification results. Experimental results show that TWSVM performs well in both digital literacy assessment and mental health identification tasks, better than other comparative models. The chi-square test results indicate a statistically significant link between digital literacy and mental health. Specifically, individuals with greater levels of digital literacy and students with higher digital literacy are more prone to having good mental health, whereas students with lesser digital literacy are more susceptible to experiencing mental health issues. This study reveals the important relationship between digital literacy and mental health among college students, providing theoretical support and practical guidance for educators and mental health professionals. Improving students’ digital literacy not only helps their academic and career development but may also positively impact their mental health and promote their overall development. Nevertheless, this study is constrained by a small sample size and a limited scope of data collection, potentially impacting the applicability of the findings to a broader population. Future studies should prioritize increasing the sample size, incorporating a wider range of sample data, and employing additional features and multi-dimensional analysis methodologies to enhance the validity and refinement of the research findings. In addition, more advanced machine learning and deep learning models can be explored to improve the accuracy and practicality of digital literacy assessment and mental health identification, providing more effective support and guidance for students’ comprehensive development.

Acknowledgment

This work was supported by the project titled “Research and practice on the evaluation of digital literacy of higher vocational students in the AI era” under Grant 2023JSJG032.

Conflicts of Interest

There are no conflicts to declare.

ORCID

Ke Bao  https://orcid.org/0000-0003-3123-2472

Junping Hu  https://orcid.org/0009-0005-1277-5545

Chun Kai Leung  https://orcid.org/0000-0002-9255-796X

Jian Yao  https://orcid.org/0009-0007-8044-5518