Online Session 1
Order	Time	ID	Full information
1	13:00-13:15	IT1033	Research on the Teaching Characteristics of Excellent Teachers in Primary School  Mathematics Dedicated Classroom Based on Video Analysis Author(s): Xinyue Cui, Yu Sun Presenter: Xinyue Cui, Yunnan Normal University, China Abstract: This study explores the teaching characteristics of excellent teachers in primary school mathematics dedicated classrooms based on systematic video analysis. Grounded in the theory of multiple intelligences, the Flanders Interaction Analysis System, and core mathematics literacy, it establishes an analytical framework that examines teaching objectives, teacher roles, language and behavior, and technology use. Four representative classes were coded and analyzed using NVivo 12 to reveal key behavioral patterns. Results show that excellent teachers demonstrate precise hierarchical teaching goals, flexible teaching roles, highly interactive communication, and efficient integration of technology. These strategies not only enhance instructional quality and student engagement but also promote educational equity by ensuring synchronous learning across main and remote classrooms.
2	13:15-13:30	IT1038	Research on Learning Behavior Recognition in Smart Classrooms Based on Artificial Intelligence Author(s): Shuangqi Li, Ran Wang Presenter: Shuangqi Li, Shanghai Normal University Tianhua College, China Abstract: With the rapid development of intelligent  education, intelligent technology dominated by artificial  intelligence is changing the traditional classroom behavior  analysis mode. Students’ classroom behavior can effectively reflect  the learning efficiency and the teaching quality of teachers, but the  accuracy of current students’ classroom behavior identification  methods is not high. Aiming at this research gap, an improved  algorithm based on multi-task learning cascaded convolutional  neural network architecture is proposed. Through the improved  algorithm, a face recognition model is constructed to identify  students' classroom behavior more accurately. Through computer  vision, natural language processing, and deep learning  technologies, this study systematically analyzes students  'classroom learning behaviors. The research developed an  integrated algorithm model combining data collection, feature  extraction, and behavioral classification, with weighted average  fusion methods determining the contribution of each component.  This provides scientific support for evaluating and analyzing  students' classroom learning behaviors. Results demonstrate that  real-time teacher analysis can reduce achievement gaps among  students with varying learning abilities. As the first experimental  study to show how real-time teacher analysis enhances student  learning, the system achieves 89.7% accuracy in accurately  identifying classroom behaviors. It also offers data-driven insights  for teachers' instructional practices and strategy adjustments,  effectively integrating artificial intelligence into education.  However, practical implementation requires attention to  educational integration pathways, ethical considerations, privacy  protection measures, and effective technology application methods  to support educational equity and personalized learning.
3	13:30-13:45	IT4450	Application of an AI-Assisted Gait Detection System for Improving Children’s Sports Interest in Preschool Physical Activity Author(s): Min Feng, Shu Hong, Liqian Liang Presenter: Liqian Liang, Shanghai Normal University Tianhua College, China Abstract: This study aimed to examine the effectiveness of an AI-Assisted Gait Detection System integrated into preschool physical activity curricula in improving children’s sports interest. A smart insole–based system with multi-dimensional plantar pressure sensors and real-time AI feedback was developed and embedded into a 16-week physical activity intervention for 100 senior kindergarten children in Shanghai, using a quasi-experimental pretest–posttest control group design. The experimental group participated in AI-assisted physical activity sessions with real-time gait monitoring and corrective guidance, while the control group followed a conventional curriculum. Results indicated no significant baseline differences between groups, whereas post-test findings showed that the experimental group demonstrated significantly higher sports participation, positive interest, and attention to physical activity, along with significantly lower negative interest compared to the control group. Within-group analyses further confirmed significant improvements only in the experimental group. Overall, the study provides empirical evidence that integrating AI-assisted gait detection technology into preschool physical education offers a data-driven and developmentally appropriate approach to promoting sports interest and active participation in early childhood.
4	13:45-14:00	IT1061	CIPP–CRITIC Dynamically Weighted and AI-Augmented Evaluation System for Front-Loaded Undergraduate Graduation Project Author(s): Shen Tao, Jiang Jin-Gang, Xiong Jiang-Long, Bao Yu-Dong, Sun Jian-Peng, Liu Chi Presenter: Jiang Jin-gang, Harbin University of Science and Technology, China Abstract: Under the guidance of New Engineering, Outcomes-Based Education (OBE), and engineering education accreditation, undergraduate graduation projects are expected to be evidence-traceable, result-comparable, and continuously improvable. To address how front-loaded graduation projects can become measurable and improvable, this study builds an evaluation system that takes the Context–Input–Process–Product (CIPP) model as the backbone and employs Criteria Importance Through Intercriteria Correlation (CRITIC) for within-dimension dynamic weighting. An AI-augmented layer is superimposed to form an integrated “evaluation–diagnosis–improvement” mechanism. Sixteen secondary indicators are identified via the Delphi method. Baseline weights are derived using the Analytic Hierarchy Process (AHP). CRITIC then performs data-driven calibration within each dimension, while fusion and smoothing preserve directional stability. A governance loop of “indicator diagnosis–expert review” ensures that indicators remain usable, current, and comparable. Case verification demonstrates stable operation, adaptive weight calibration to real data, and strong interpretability. The system provides data support and actionable governance levers for building replicable and scalable quality-assurance pathways, advancing the institutionalized implementation and high-quality development of front-loaded graduation projects.
5	14:00-14:15	IT2135	AI Literacy Test for University Students: Design Principles, Content Validation, and  Pilot Testing Author(s): Lina Zhang, Shuhan Zhang Presenter: Zhang Lina, Macao Polytechnic University Abstract: The increasing integration of AI into higher education significantly heightens the need for developing robust instruments to measure university students' AI literacy. However, there remains a lack of fully validated, objective tests specifically for summative assessment appropriate for higher education students. To bridge this gap, we designed the AI Literacy Test for Higher Education students (AILT-HE). The test was developed using the Evidence-Centered Design (ECD) framework to measure six constructs, comprising an initial set of 53 items. In the validation phase, content validation involved expert judgment and cognitive interviews, which guided the initial test revision. Subsequently, a pilot test was conducted at an undergraduate institution (N=69). Psychometric analyses indicated that the test demonstrated appropriate difficulty levels and reliability. These findings guided the refinement of the instrument. Finally, implications for future research and applications in higher education were discussed.
6	14:15-14:30	IT2168	Real-Time Diagnosis and Intervention of Teaching Behaviors Driven by Machine Learning in New-Engineering Classrooms Author(s): Wei Guo, Zeyu Yan Presenter: Wei Guo, Wuhan Institute of Technology, China Abstract: China’s New Engineering programming courses lose one fifth of first year students because instructors cannot spot the seven-minute silent confusion cycle that unfolds while they face the IDE. We trained an edge compute system on thirty thousand labelled frames extracted from one hundred twenty hours of video recorded in ten courses with two hundred eighty four students. A pruned YOLOv8 Transformer ensemble running on a four gigabyte Jetson Nano detects eight micro behaviors such as keyboard silence head pose drift and IDE error bursts at twenty five frames per second within forty milliseconds. When the posterior probability of confusion or distraction exceeds nought point seven for three consecutive frames the board flashes a three level cue on the instructor s secondary monitor and pushes a four line AST matched code hint into each student s IDE. An A B A experiment with one hundred eight novices showed immediate gains. Eye movement entropy fell from three point nine zero to three point one eight bits and quiz accuracy rose from seventy two point seven per cent to eighty four point six per cent. Four weeks after withdrawal retention remained seven point three percentage points above baseline. The bill of materials is two hundred seventeen euros per classroom. The dataset and Jetson image are released under MIT License for three hundred room scale up.
7	14:30-14:45	IT4432	Automatic Grading of Chinese as a Second Language Reading Texts Based on Multi-LLM Weighted Voting Author(s): Xinxin Han, Juan Xu Presenter: Xinxin Han, Beijing Language and Culture University, China Abstract: To address the issues of unstable prediction and discriminatory biases of single LLMs in the automatic grading of Chinese as a Second Language reading texts, this paper proposes a text grading method based on multi-LLM weighted voting. This research constructs a decision fusion framework for heterogeneous LLMs, introduces a conditional weight calibration mechanism to dynamically allocate weights according to LLM performance across different levels, and adopts a weighted median strategy to suppress outlier predictions. Experimental results show that the proposed method is superior to single LLMs in terms of Accuracy, Quadratic Weighted Kappa (QWK), and Mean Absolute Error (MAE), verifying the effectiveness of multi-LLM collaboration in improving grading accuracy and system robustness.
8	14:45-15:00	IT4447	A Novel BiLSTM Hybrid Model Fused with Speaker Information and Multi-Module Collaboration for Classroom Dialogue Text Classification Author(s): Qingtang Liu, Junji Xiao, Xinyu Jiang, Ruyi Jiang, Xinqian Ma, Kun Huang Presenter: Junji Xiao, Central China Normal University, China Abstract: Traditional classroom dialogue analysis relies on manual coding, which does not adapt to the characteristics of local n-gram features, strong context dependence and speaker ID in dialogue classification, which affects the discrimination ability and classification performance of the model. This study proposes a BiLSTM-based hybrid text classification model according to the characteristics of classroom dialogue. The model introduces CNN in the encoder to capture local n-gram features, and uses BiLSTM to get the global context. Sentence-level information is aggregated via multi-strategy pooling, and a MLP classification head is employed to enhance the expressive capacity of nonlinear. In addition, speaker is integrated into the input as a prefix to make use of the speaker information in the classroom dialogue, and verifies it under the speaker on/off experiment. The experimental results show that the Macro-F1 model improves by 2.62% on IRE (S off); On IRE (S on), Macro-F1 reached 90.29%. Ablation experiments verified the synergistic improvement effect of each module. On the three public datasets, the model Macro- F1 improved by 1.53%, 4.58% and 3.86%. The research shows that the proposed method provides an effective scheme for the automatic analysis of classroom dialogue.
9	15:00-15:15	IT1049	ode2vec-Based Graph Embedding for Deep Learning Graduation Project: Advanced Management Author(s): Jiang Jin-Gang, Fu Kang, Shen Tao, Bao Yu-Dong, Kang Fu-Wei, Dong Jing-Hao Presenter: Jiang Jin-gang, Harbin University of Science and Technology, China Abstract: Graduation projects serve as a crucial assessment of undergraduate students' academic achievements over four years, playing an irreplaceable role in talent cultivation. Course instruction, as the primary vehicle for imparting theoretical knowledge, directly influences students' ability to effectively apply learned theories to solve complex engineering problems when undertaking graduation projects. However, current teaching practices face challenges including disconnect between theory and practical application, poor alignment between course content and graduation project requirements, inadequate guidance system coordination, and insufficient personalized instruction. This paper employs the Node2vec algorithm to optimize the course alignment mechanism for advancing graduation projects, ensuring tight integration between course content and design requirements. It establishes a topic selection model that enhances quality by incorporating innovative achievements such as outstanding competition awards, outcomes from undergraduate innovation and entrepreneurship training programs, converted faculty research project results, and completed enterprise-demand projects. This study provides a scalable implementation scheme for cultivating application-oriented, innovative, and interdisciplinary talents in the new era.