Online Session 2
Order	Time	ID	Full information
1	13:00-13:15	IT2114	A Study on Perceptions of Application and Risk Perception of Artificial Intelligence in Primary and Secondary Education Author(s): Yue Zhao, Jiachao Wei, Minxi Zhang Presenter: Yue Zhao, Nanning Normal University, China Abstract: Grounded in risk perception theory, this study systematically investigates differences in risk perception regarding artificial intelligence (AI) applications among educational stakeholders. It employs questionnaire surveys and empirical analysis based on 952 valid responses collected from educational stakeholders in Guangxi Province, with a primary focus on teachers and students. Findings reveal significant cognitive discrepancies in current AI educational applications: overall characterised by "high usage rates, high ambiguity, and low concrete cognition", with over half of respondents exhibiting vague risk perceptions and 90.2% unable to clearly articulate risk types; A "cognitive misalignment" exists among core groups: pupils exhibit "high vigilance, low ambiguity", while teachers display "high ambiguity, low vigilance". Usage experience emerges as a key variable influencing risk perception clarity, highlighting that the digital divide has extended to the cognitive level. The study indicates that promoting AI education must transcend technical training, shifting towards "cognitive collaboration" to strengthen risk consensus-building and align teacher-pupil perceptions, thereby fostering rational technological application
2	13:15-13:30	IT2131	Online Course UX Evaluation via Sentiment Analysis and Scale Validation Author(s): Haoyu Zhang, Mei Wang, Shuai Wang, Jiuyan Zhou, Binrui Jiang Presenter: Mei Wang, Sichuan Tourism University, China Abstract: User experience (UX) plays a critical role in the quality and effectiveness of online teaching. This study analyzes user comments from four computer science courses on the China University MOOC (CUM) platform using Natural Language Processing and sentiment analysis (NLP) to identify key user concerns. The courses are categorized into foundation courses and major courses, and topic modeling is performed separately for each category. LDA-based topic modeling identifies five major UX-related themes, and a comparative sentiment analysis is further conducted to examine emotional differences between foundation and major courses across these themes. Based on the extracted themes and sentiment insights, an online course UX scale is developed with those dimensions: Interactive Experience (IE), Content Quality (CQ), Learning Outcome (LO), Teaching Quality (TQ), and Teaching Support (TS). A two-stage survey was conducted using a 7-point Likert scale, and the scale’s reliability and validity were validated through exploratory and confirmatory factor analyses. The results reveal significant relationships between learners’ emotional experiences, learning outcomes, and learning motivation. This study provides empirical evidence for improving online course design and demonstrates the effectiveness of integrating sentiment analysis with UX scale development in online teaching research.
3	13:30-13:45	IT2176	Tripartite Evolutionary Game and Governance Strategies for Integrating Generative AI into Higher Education Teaching: Based on Strategy Trade-off Analysis of schools,Teachers,and Students Author(s): Lin Wang Ying Jiang Presenter: Lin Wang, Heilongjiang International University, China Abstract: Abstract—The rapid development of generative AI is driving the transformation of teaching models in higher education, but it also raises issues such as increased governance costs, heightened risks of academic misconduct, and adaptation difficulties for teachers and students. While current research often surveys user willingness, few studies analyze the systemic conflicts of interest among stakeholders. To address this, this paper builds a tripartite evolutionary game model involving schools, teachers, and students. Using Python for numerical simulations, we analyze how these parties adjust their strategies. Results indicate that while initial willingness affects evolutionary speed, the final stable state depends on specific parameter constraints. Simulations reveal that breaking the current deadlock requires three simultaneous conditions: schools must lower management costs via technology while maintaining risk sensitivity; teachers must receive sufficient incentives to compensate for the costs of learning new technologies; and students require a dual "tech + manual" supervision system to curb misconduct. Accordingly, we propose governance strategies such as lowering technical barriers and establishing collaborative supervision.
4	13:45-14:00	IT3226	The relationship between college students' GenAI usage patterns and learning engagement: a latent profile analysis Author(s): Siqi Peng, Kejun Zhuang, Baoxun He Presenter: Siqi Peng, China West Normal University, China Abstract: Generative Artificial Intelligence (GenAI) has emerged as a significant tool in college students' learning processes. The patterns of its usage reflect individual behavioral characteristics in engaging with AI technologies, making it essential to comprehensively understand these patterns and provide appropriate guidance. To this end, this study conducted an online survey of 470 Chinese college students using the College Students' GenAI Usage Questionnaire and the Learning Engagement Scale. Latent profile analysis (LPA) was employed to identify latent classes of GenAI usage patterns. Subsequently, the robust three-step approach was employed to examine the influence of demographic variables and other factors on these patterns. Finally, the Bolck-Croon-Hagenaars (BCH) method was applied to investigate differences in learning engagement across distinct usage pattern groups. The results indicate that: a. College students’ GenAI usage patterns are best classified into four distinct categories: "peripheral experimenters, " "tool-dependent users, ""cautious critics, "and "normative deep users. " b. Academic year and GenAI proficiency significantly predict category membership, with GenAI proficiency demonstrating the strongest predictive effect. c. Significant differences in learning engagement were found among students with different GenAI usage patterns. These findings transcend the limitations of traditional variable-centered approaches and provide direction for developing targeted intervention strategies.
5	14:00-14:15	IT3278	UniIP-RiskNet: A Transformer-Based Large Language Model for Intellectual Property Contract Risk Early Warning in Higher Education Scenarios Author(s): Xue Zhang Presenter: Xue Zhang, Guangzhou College of Applied Science and Technology, China Abstract: With the increasing complexity of university– industry collaboration and technology transfer, intellectual property (IP) contracts have become a critical instrument for governing research outputs in higher education institutions. However, the manual review of such contracts is time-consuming, error-prone, and difficult to scale, particularly when contracts contain heterogeneous clauses related to academic publication rights, background knowledge protection, and compliance with institutional policies. To address these challenges, this study proposes UniIP-RiskNet, a Transformer-based hybrid and explainable framework for early warning of IP contract risks in higher education scenarios. The proposed approach performs clause-level risk analysis by integrating deep semantic representations learned from a legal-domain Transformer with rule-based red-flag enforcement and knowledge graph consistency validation, enabling accurate, interpretable, and policy-aware risk detection. In addition, a retrieval-augmented generation module is introduced to provide actionable remediation suggestions grounded in institutional clause templates and policy documents, supporting contract optimization beyond risk identification. Experimental results demonstrate that UniIP-RiskNet consistently outperforms representative baseline methods across multiple evaluation metrics, particularly in specialization, effectiveness, and accuracy. These findings confirm the effectiveness and practical value of the proposed framework for intelligent IP contract governance in higher education institutions.
6	14:15-14:30	IT3320	Exploring the Relationship between AI Trust and AI Literacy among Chinese College Students Author(s): Xiao Zhou, Zhixiang Pan, Liu Li Presenter: Xiao Zhou, Guangdong University of Science and Technology, China Abstract: With the rapid integration of artificial intelligence (AI) into higher education, understanding factors that shape students’ AI literacy has become increasingly important. This study investigates the relationship between AI trust and AI literacy among Chinese college students and examines the extent to which AI trust predicts AI literacy. Using a questionnaire survey, data were collected from 485 college students in Guangdong and analyzed using partial least squares structural equation modeling (PLS-SEM). The results reveal a significant and positive relationship between AI trust and AI literacy. Further analysis indicates that AI trust demonstrates moderate explanatory power and strong predictive relevance for AI literacy, highlighting its substantial role in students’ AI competence development. These findings suggest that trust-related beliefs toward AI are important psychological antecedents of AI literacy. The study contributes to the literature by linking AI trust with AI literacy and offers practical implications for designing educational interventions that foster both students’ trust in AI technologies and their AI-related competencies.
7	14:30-14:45	IT3339	The Ethical Architecture of Assessment: Negotiating Teacher Value-Judgments in Generative AI Contexts Author(s): Grace T. Flores, Kheven D. Guyo Presenter: Grace Flores, Caraga State University, Philippines Abstract: The rapid integration of generative artificial intelligence (GenAI) into higher education has disrupted traditional assessment practices, raising concerns regarding authorship, originality, fairness, and academic integrity. While existing discourse largely centers on detection technologies and regulation, limited research examines how teachers construct and negotiate ethical judgment in AI-mediated assessment contexts. Addressing this gap, this study develops a grounded theoretical framework, the Ethical Architecture of Assessment, based on semi-structured interviews with thirteen higher education educators across disciplines. Through inductive coding and reflexive thematic analysis, five interconnected processes emerged: (1) suspicion as ethical vigilance, (2) guidance as ethical mediation, (3) adaptation versus resistance as ethical negotiation, (4) fairness and integrity as anchoring principles, and (5) shifting assessment modalities as ethical innovation. Together, these processes form a relational system through which teachers balance authenticity, integrity, and pedagogical responsibility. Interpreted through procedural, interactional, and educational fairness dimensions, the findings demonstrate that professional judgment in AI-integrated classrooms is ethically structured rather than technologically reactive. Although grounded in qualitative depth within a single institutional context, broader empirical validation across disciplines and institutions is needed. This study contributes an original, teacher-centered model reframing AI integration as an ongoing ethical negotiation embedded in professional practice.
8	14:45-15:00	IT3386	AI-Empowered Heterogeneity Response in Remote Classrooms: A Qualitative Study on the Mechanism of Human-AI Collaborative Assessment and the Construction of a Dynamic Ecological Model Author(s): Bin Chen, Hui Sun Presenter: Bin Chen, Jinhua Open Universtiy, China Abstract: As remote classrooms have demonstrated a pivotal role in promoting educational equity and supporting lifelong learning, how to effectively respond to learners’ heterogeneity in large-scale teaching has become a core challenge in improving educational quality. Addressing this issue, this study adopted the grounded theory method to explore the dynamic process of how teachers and AI collaboratively respond to student heterogeneity in remote classrooms. We conducted semi-structured interviews with 33 frontline teachers experienced in AI-assisted remote teaching. Through the hierarchical coding of raw data, the study constructed an AI-assisted heterogeneity response strategy model for remote classrooms. The model reveals that effective heterogeneity response is a dynamic circular system with “Human-AI Collaborative Assessment and Judgment” as the value conversion hub: Intelligent technology perceives and diagnoses students’ differences through multimodal data; on this basis, teachers integrate professional experience and situational insights to complete the translation from data to educational understanding; furthermore, through multi-dimensional and dynamic intervention strategies covering content, interaction, process, and structure, the assessment and judgment are put into teaching practice. The entire process relies on a systematic digital ecology as the supporting environment, with the fundamental goal of promoting students’ subjective development such as thinking autonomy, sense of identity meaning, and comprehensive agency. This study not only provides a systematic theoretical framework for understanding how teachers, students, and technology collaboratively respond to heterogeneity in remote classrooms in the intelligent era but also offers practical paths for educational practitioners to improve teaching quality in terms of heterogeneity diagnosis, strategy application, ecological cultivation, and evaluation reform.
9	15:00-15:15	IT4465	Research on Multi-Dimensional Learning Behavior Modeling and Grade Prediction  Based on Transformer Author(s): Yan Zhou, Jun Zhou, Xin Gao Presenter: Yan Zhou, Wuhan Business University, China Abstract: Traditional teaching evaluation methods, which primarily rely on final examination scores, struggle to comprehensively capture students' multi-dimensional competency development in blended learning environments. To address this limitation, this paper proposes a teaching evaluation prediction system based on the Transformer model. The system integrates student data across five dimensions: online learning behavior, classroom participation, assignment scores, project collaboration, and final examination grades. By employing a multi-head self-attention mechanism to model the nonlinear dependencies among these features, the system achieves accurate prediction of students' academic performance. Experiments were conducted using real learning data from 100 students, partitioned into training, validation, and test sets at a ratio of 7:1.5:1.5. Results demonstrate that the model achieves a prediction accuracy of 88% on the test set, with a mean squared error of 0.022. The correlation coefficients for both training and test sets approach 1, validating the model's strong fitting capability and generalization performance. Furthermore, the system provides teachers with real-time feedback on student learning progress and supports personalized teaching decision-making, facilitating a shift from outcome-oriented to process-oriented teaching evaluation. This research demonstrates that the Transformer-based evaluation prediction model can effectively enhance the precision of teaching assessment and provide data-driven technical support for personalized learning interventions in blended teaching environments.