Research collaboration is important for STEM doctoral students’ academic socialization, research productivity, and career development. However, existing research has paid limited attention to its dynamic evolution over the doctoral journey. This study explored how research collaboration developed among Chinese STEM doctoral students by integrating institutional logics theory with a temporal network analysis. Drawing on longitudinal data from 20 final-stage STEM doctoral students at two top research universities in China, the study identified three cumulative phases in their collaboration networks: dependency, diversification, and realignment. The findings showed that these students’ research collaboration was a temporally embedded and institutionally mediated practice, shaped by shifting institutional logics of academic training, research production, and career development across different stages of doctoral study. Moreover, students’ ability to diversify and realign their networks was unevenly shaped by the nature and extent of support provided by supervisors and institutions. The study recommends that hierarchical doctoral programs should foster equitable collaboration by offering stage-specific support, encouraging diversified mentorship, and adopting inclusive evaluation policies. While situated in the Chinese context, this study offers a novel temporal institutional logics perspective that enhances our understanding of the evolution of doctoral research collaboration internationally.

Programming education in primary school is vital for nurturing future-ready talents, yet primary school students often struggle with self-regulated learning (SRL), particularly in resource utilization and strategy regulation. Although human-generative AI (GenAI) collaborative programming learning might have the potential to enhance personalized programming education, GenAI's interplay with SRL processes remains underexplored. To address this gap, this study first proposed a SRL Behavior Analysis Framework for human-GenAI collaborative programming learning environments and then examined SRL behaviors of a group of sixth-grade students (n = 36) in such an environment using this framework, along with various learning analytics methods including cluster analysis, descriptive statistics and lag sequential analysis. The analysis yielded the following results: (1) Based on their learning performance, primary school students were identified as three distinct clusters: programming specialized unit (PSU), high performance unit (HPU), and low performance unit (LPU). (2) Regarding SRL behaviors, students prioritized selfcontrol (65.8 %), followed by self-observation (19 %), task analysis (12.1 %), and behavior stagnation (3.2 %). (3) Students in PSU and HPU consistently adopted goal-oriented SRL strategies, whereas students in LPU exhibited passive dependence and fragmented strategy use. GenAI's facilitative effect in supporting learning correlated with users' SRL capabilities. (4) Students in PSU and HPU exhibited frequent transitions between SRL behaviors, whereas students in LPU had insufficient ability to switch strategies when facing programming difficulties. Based on these findings, this study proposed four forward-looking design recommendations: effectively integrating GenAI with the programming environment, utilizing multimodal data and AI for learning assessment and feedback, building a cluster-driven early warning mechanism, and conducting dynamic SRL analysis and guidance based on fine-grained time-series data.

Given that ChatGPT can write essays and has become an accessible tool for students, it is important to understand how it may transform the nature of writing and what such a transformation implies for the education of the young generation. While most existing studies have focused on the quantitative effects, limited attention has been paid to the underlying writing processes and students' behavior patterns. Without a deeper understanding of how students engage with human-ChatGPT collaborative writing (HCCW), it is challenging to ensure its safe and effective integration into educational contexts. To address this gap, this study utilized a mixed-methods approach to explore high school students' behavior patterns, performance, and perceptions in HCCW. A sample of 53 high school students engaged in two writing tasks: traditional individual writing and HCCW. Data were collected through screen recordings capturing students' real-time interactions with ChatGPT, essay assessments, questionnaires, and semi-structured interviews. The findings are as follows: (1) There are three typical behavior patterns observed among the participants in HCCW, including co-writing with ChatGPT, instructing ChatGPT to directly generate essays, and writing independently without ChatGPT's assistance. (2) HCCW can significantly improve essay quality, though it requires more time than traditional individual writing. (3) Students who performed well in traditional individual writing also tended to excel in HCCW. HCCW particularly benefited students with moderate and poor writing abilities, bringing their writing performance closer to that of their more adept peers. (4) More than 60% of participants recognized ChatGPT's value as a writing tool, appreciated the quality of the final product, and expressed a willingness to continue using it for writing. Only 20% of the students felt a sense of authorship over the collaboratively produced essays. HCCW also raises ethical concerns, including the emotional toll, loss of voice, and over-reliance on the tool. These findings deepen our understanding of the nuances of HCCW and provide practical implications for guiding students to use AI tools ethically and productively. Additionally, this research can guide educators in effectively navigating the transformative challenges of generative AI in writing education.

Computational thinking (CT) is crucial for enhancing students’ complex problem-solving abilities in the intelligent era. The emergence of generative artificial intelligence (GenAI) is profoundly transforming the global educational landscape and demonstrating significant potential for promoting personalized learning. However, the literature offers varied results on the effectiveness of using GenAI to cultivate students’ CT. This study comprehensively investigated the effects of GenAI on students’ CT and the role of moderating factors, integrating 45 effect sizes from 25 empirical studies published between 2022 and 2025. A theoretical framework of factors influencing students’ CT was proposed based on activity theory, and the moderating factors included educational level, region, intervention duration, teaching mode, interaction mode, role setting, and feedback type. The results indicated that GenAI had a significant overall positive effect on students’ CT development. Specifically, the largest effect size was computational practice, followed by computational concept and computational perspective. Furthermore, the analysis revealed that region, teaching mode, and interaction mode had significant moderating effects. Based on these results, this study offers targeted implications across the dimensions of theoretical foundation, educational practice, and technological development, providing empirical evidence for implementing GenAI teaching and developing GenAI tools to cultivate students’ CT.

ChatGPT, an AI-driven chatbot capable of generating code and providing timely, personalized feedback, offers new opportunities for supporting programming learning among younger learners. This study employed a quasi-experimental design to examine the effects of Self-Directed Programming (SDP) and ChatGPT-Facilitated Programming (CFP) learning modes on primary learners’ programming knowledge, skills, and attitudes. In addition, learners’ programming behaviors were explored within the CFP learning mode. A total of 71 primary learners were randomly assigned to either the CFP or SDP class. Learners in the SDP class engaged in self-directed programming activities, whereas learners in the CFP class learned programming through a ChatGPT-facilitated learning platform, CodewithAI. A mixed-methods approach was adopted to collect and analyze multidimensional data. The results revealed a significant difference in programming knowledge performance between the CFP and SDP classes. Behavioral analysis indicated that primary learners in the CFP class frequently sought assistance from ChatGPT when encountering programming challenges, particularly by requesting explanations of code and error messages. In addition, primary learners in the CFP class reported relatively high levels of confidence compared with those in the SDP class. Based on these findings, this study offers implications for the design of primary programming instruction and the development of AI-supported instructional tools. By providing fine-grained empirical evidence, this study contributes to the field of the effectiveness of ChatGPT-facilitated programming learning for primary students.

Creative thinking is a core competency for cultivating innovative talents in the 21st century. Design-based learning (DBL), an emerging instructional model, emphasizes iterative project refinement through student-driven inquiry and design, thereby fostering the construction of meaningful knowledge. Although previous research has demonstrated the potential of the DBL model to enhance students’ creative thinking, clear and systematic procedures for implementing DBL in middle school contexts remain insufficiently specified. To address these gaps, this study first constructed a preliminary DBL model aimed at fostering creative thinking. The final DBL model was empirically derived from two rounds of practice in the STEM course at L Middle School in China. A mixed-method approach was employed to examine how the proposed DBL model supported students’ learning performance, digital works, creative thinking, and learning experiences across the two rounds of activities. The results demonstrated a marked improvement in the creativity of students’ digital work, accompanied by progressive improvement in learning performance and a significant increase in creative thinking skills. Furthermore, student feedback indicated a strong preference for the DBL model and a belief in its effectiveness in nurturing their creative thinking. This case study provided both a validated practical instructional model and conceptual insights for developing DBL to cultivate students’ creative thinking in K-12 education.

Clinician-scientists are widely regarded as critical for translational medicine, yet many systems struggle to sustain a stable clinician-scientist workforce. Existing research often assumes a single linear identity pipeline and focuses on Western MD-PhD programs, with limited attention to heterogeneous trajectories in other high-intensity contexts. In China, rapid expansion of dual-role PhD programs in elite “Double First-Class” medical universities has created dense expectations in both clinical service and research. How medical PhD candidates position themselves within these dual demands, and how distinct clinician-scientist identity pathways emerge, remains insufficiently understood. A qualitative multi-case study was conducted with 30 medical PhD candidates from 30 top-tier medical universities in North and East China. All participants held a medical degree and were engaged in both clinical and research training. Semi-structured interviews explored motivations, professional self-definition, experiences of dual-role conflict, perceptions of evaluation systems, mentorship arrangements, and access to institutional support. Transcripts were analyzed through inductive coding followed by ideal-typical analysis to construct comparative pathways of clinician-scientist identity development. Four ideal-typical pathways were identified. The Integrated Dual-Identity type described clinical and research roles as mutually informative and actively sought synergy across settings. The Clinician-First type regarded the PhD primarily as an instrument for clinical career advancement and tended to meet only minimum research requirements. The Research-First type oriented mainly toward a scientific career and substantially reduced clinical engagement during training. The Ambivalent type reported persistent tension between domains, oscillating commitments, and marked psychological strain. Evaluation criteria, mentor configurations, organization of clinical and research time, and availability of institutional resources systematically steered trainees toward particular pathways and rendered some trajectories more fragile than others. Clinician-scientist identity formation among medical PhD candidates follows multiple pathways. The four ideal types provide an analytic framework for understanding how local structures channel trainees toward integrated dual identities, single-role orientations, or sustained ambivalence. Programs that combine flexible evaluation systems, coordinated clinical and research mentorship, structured protection of time for both roles, and targeted support for ambivalent trainees may strengthen the clinician-scientist pipeline in China and offer transferable design principles for similar high-pressure training environments worldwide.

Against the background that artificial intelligence (AI) technology has profoundly reformed higher education, the teaching of AI courses in universities is facing structural challenges such as outdated content, insufficient computing resources, weak ethical education, and monotonous assessment. Taking the u201CIntroduction to Artificial Intelligenceu201D course at Zhejiang University as an example, this paper analyzed the design and implementation effectiveness of its integrated theory-practice teaching model of u201Cclassroom teaching + platform trainingu201D supported by a new-generation science and education platform. The research showed that this model effectively promoted the formation of studentsu2019 ability structure of u201Ccognitive spiral developmentu201D through dynamic content updating, progressive practice, full-process ethical guidance, and multi-dimensional evaluation integration. Empirical results indicated that platform training significantly improved studentsu2019 practical abilities and learning outcomes, while course perception and environmental support had a significant positive impact on AI literacy. Based on multi-source feedback, this paper put forward optimization strategies from four aspects: course content, platform functionality, evaluation mechanisms, and governance frameworks, aiming to provide an operable reference model for AI curriculum reform and the cultivation of interdisciplinary talents in universities.

Background Current human-GenAI collaborative programming predominantly defaults to Forward Engineering Pedagogy (FEP), where novices frequently encounter debilitating cognitive load and inefficient exploratory cycles, particularly those lacking robust self-regulated learning (SRL) skills. Conceptually, Reverse Engineering Pedagogy (REP) is hypothesized to mitigate this by providing a functional anchor that explicitly scaffolds the forethought and self-reflection phases of SRL. To validate this mechanism, this study investigates how FEP and REP differentially shape students' SRL and cognitive engagement. Because these pedagogies fundamentally alter the temporal dynamics of learning, process-oriented methodologies are conceptually necessitated. An empirical investigation spanning eight weeks compared the FEP and REP instructional models among 70 primary school participants. Multimodal datasets comprising screen recordings and dialogue transcripts were evaluated utilizing descriptive statistics, chi-square tests, lag sequential analysis (LSA), and epistemic network analysis (ENA). Results Results reveal three principal findings. First, REP significantly improved students' SRL distributions, enhancing self-observation and mitigating unproductive trial-and-error. Second, LSA demonstrated that REP cultivated highly adaptive SRL behavioral trajectories and strategic flexibility. Third, ENA confirmed that REP deepened students' cognitive engagement, facilitating a critical transition from superficial information retrieval to profound knowledge construction. Conclusion Ultimately, this research theoretically and empirically validates REP as a robust instructional scaffold in AI-enhanced environments. By offloading initial cognitive constraints, REP reconfigures the interaction dynamic, transforming GenAI from a superficial shortcut into a catalyst for deep cognitive engagement. Subsequent longitudinal investigations should consider integrating targeted interventions focused on prompt engineering.

As artificial intelligence (AI) continues to transform global education, the development of AI literacy has become a pressing priority. However, current initiatives often approach AI literacy from local or national perspectives, overlooking the importance of a broader, global context. In cross-cultural learning, immersive virtual reality (VR) environments based on 360-degree video technology offer authentic experiences; however, these are typically asynchronous, limiting learners' opportunities for real-time interaction, feedback, and support. Such limitations may hinder the effective development of global AI literacy, innovation, and entrepreneurship. To address this challenge, the present study designed and implemented AI-powered interactive pedagogical agents (PAs) within immersive VR environments. These agents provided real-time, context-sensitive support during cross-cultural learning activities, aiming to enhance engagement and promote global AI literacy, innovation, and entrepreneurship. A total of 68 students from China, Northern Cyprus, and Kazakhstan participated in the study. With the assistance of PAs, students created cross-cultural scripts and recorded corresponding 360-degree videos, which were then exchanged among participants. Using VR headsets, students viewed immersive learning content on AI concepts and their applications across diverse cultural contexts, engaging in verbal interactions with PAs throughout the viewing experience. Following this, students reflected on their learning experiences under the guidance of PAs. A mixed-methods research, comprising pre- and post-test questionnaires, interviews, and reflection reports, demonstrated statistically significant gains in students' global AI literacy, innovation, and entrepreneurship following the intervention. Participants also reported high levels of perceived credibility and immersion with the PAs in VR environments. Based on these findings, the study offers recommendations to inform future practices in global AI literacy education.

Blended Synchronous Classrooms (BSCs) demonstrate advantages in advancing educational equity. However, few studies have focused on Q&A interactions in BSCs. Unlike traditional single-class settings, BSCs require teachers to simultaneously manage interactions with students from two classes. To address the gap, this study proposed a coding framework for Q&A behaviors grounded in the Revised Bloom's Taxonomy. This study analyzed 40 BSC lessons using various learning analytics methods, including Lag Sequential Analysis, Social Network Analysis, and K-Means Clustering Algorithm. The data analysis yielded the following results: (1) Q&A participants: Teachers prioritized interactions with the students in their own classes. (2) Cognitive levels: Most interactions remained at the Understanding level. (3) Transition patterns: Urban teachers exhibited sequential, logic-driven questioning progressions, whereas rural teachers displayed fragmented transitions confined to Understanding-level loops with weak high-low cognitive linkages. (4) Key Q&A behaviors: Urban teachers demonstrated the use of question categories at higher cognitive levels, while rural teachers only used question categories at lower cognitive level. This exploratory study represented a systematic investigation into the quality of teacher-student Q&A behaviors within the urban-rural BSCs context. Finally, the study proposes actionable strategies for optimizing BSCs' Q&A design and enhancing dialogic teaching quality in distance educational environments.

Many Chinese families are facing a caregiving crisis, where the spillover of risks causes internal family problems to evolve into broader social concerns. Since the community serves as the primary living environment, it acts as an optimal site to carry out caregiving activities. Based on a qualitative study in Northeast China, this article explores a welfare cooperative production model of community care for 'one old and one young'. The findings indicate a rising demand for community care but with an insufficient supply, which has primarily resulted from family dysfunction where the double burden of overlapping needs for elderly and young care as well as work-family conflicts plunged families into caregiving crises, with risks spilling over into the community. However, constrained by limited autonomy and a lack of social trust, community care services encountered a bottleneck. The community in this study has searched out a welfare cooperative production model, including vertical empowerment and horizontal collaborative cooperation, which facilitates two key transitions: from government-led governance to collaborative governance and from technical co-production to value co-creation. Nevertheless, in the later stages of cooperative production, the model faced issues of goal deviation and value co-destruction, where the focus of teamwork shifted towards political propaganda, and the elderly care service station faced financial deficits. Moving forward, family-centred care policies should be developed to achieve full life-cycle development and enhance family resilience. Additionally, social work teams and digital technology should be introduced to empower the cooperative production and to ensure the stable provision of care services.

Video apps are widely used by children worldwide. However, concerns have been raised about children's personal data disclosure and privacy risks during their consumption of video apps. This study designed and validated a conceptual model for children's continuance intention to use video apps based on privacy calculus theory and Teen Online Safety Strategies (TOSS) parental mediation framework. A cross-sectional survey of 321 Chinese middle school-aged children (aged 12-14) was conducted to validate the model. Our findings revealed that children's perceived risk had a minimal association with their perceived value. Children's perceived benefit was positively associated with their perceived value, which was positively associated with their continuance intention to use video apps. Trust also contributed to children's continuance intention to use video apps, slightly positively moderated by restrictive parental mediation. This study provided a better understanding of children's perception of privacy, trust and value in consuming video apps and identified the potential role of related parental mediation. To mitigate children's privacy risks and to enhance children's digital literacy as well as digital autonomy, we proposed some recommendations to video app designers, parents and educators. The study appeals to a more transparent and child-centred approach in designing video apps, a mixed parental mediation for facilitating children's video app usage, and educators' empowering children's autonomy in online privacy protection by improving children's digital skills and competencies.

Several studies have shown that computer-supported argument visualization (CSAV) tools improve students' argument analysis skills. However, instructional principles for designing CSAV tools to improve students’ argument analysis have been inadequately investigated. This paper examines the effect of a systematically designed CSAV tutorial on promoting students' argument analysis and writing skills in a non-English major course. The tutorial's design follows Merrill's (2023) First Principles of Instruction model. The study adopts a single-group pretest-posttest design, with participants consisting of 41 sophomores enrolled in the Modern Education Technology course at a Chinese university. In the pretest, students were instructed to (1) identify the components of the argument structure in a pre-assigned short text with two opposing views (argument analysis) and (2) write an argumentative essay. After receiving the designed tutorial lessons for seven weeks (2 hours and 45 minutes each week), they were post-tested on extracting the argument structure from a given text and demonstrating it in their argumentative essays. Participants scored significantly higher in argument analysis and argumentative writing in the post-test compared to the pre-test. These findings suggest that a systematic design of a CSAV tutorial improves students’ skills in argument analysis and writing argumentative essays. Instructional design principles that enhance students' argument analysis and writing skills are discussed.

Cultivating the digital literacy of vocational education normal students is a fundamental link in promoting the digital teaching reform of vocational education. Taking the vocational education normal students majoring in finance and accounting of Lingnan Normal University as the research object, this paper investigates and analyzes the current situation of the digital literacy level of normal students. The results show that the digital literacy of finance and accounting normal students is generally at a medium to high level, and their digital social responsibility literacy is good. The higher the grade of normal students, the higher their digital literacy level. This indicates that the implementation of the school's "New Normal Education + Education Informatization 2.0" and "Three Transformations and Four Modernizations" peak construction project has achieved certain results. The "Digital Technology knowledge and skills", "digital application" and "professional development" qualities of finance and accounting vocational education normal students can basically support their learning activities as college students, but are not sufficient to support their teaching activities as teachers. However, finance and accounting vocational education normal students have a good attitude and confidence in learning new digital technologies. In this regard, it is urgent to build a complete knowledge system of digital technology, integrate digital literacy education with professional education, and create diversified digital literacy education scenarios to improve the digital literacy level of normal students in finance and accounting vocational education.

Generative Artificial Intelligence (AI) is steadily gaining prominence in higher education and brings about huge impact on college students’ daily life. However, limited studies paid attention to college students’ use behavior of generative AI and its influencing factors. The study aimed to explore this issue by adopting an extended Unified Theory of Acceptance and Use of Technology (UTAUT) model. Two generative AI related variables, named “novelty value” and “perceived humanness,” were added to the UTAUT model, and the potential moderating effects of gender, grade and major were also considered in the model. The model was validated by collecting data from 1,190 college students at a Chinese university. Results indicated significant positive correlations among performance expectancy, effort expectancy, novelty value, social influence, facilitating conditions, and behavioral intention. Additionally, facilitating conditions and behavioral intention significantly influenced use behavior, while perceived humanness had no significant impact on behavioral intention. Moderating variables like gender and grade significantly affected the acceptance of generative AI, whereas major did not. The findings provided nuanced insights to advance the practical application of generative AI. Considering limited sample in the study, future research may encompass diverse demographics across multiple countries to enable cross-cultural comparisons.