Chengliang Wang

Chengliang Wang

Research highlights

Programming education and learning environments

Journal of Educational Computing Research 2025

MetaClassroom: A New Paradigm and Experience for Programming Education

Figure 2: the original overview of the MetaClassroom learning environment, with its teaching spaces and interactive stations.

Learning environment

Scene 6 from Figure 4: the virtual classroom with a programming blackboard, teaching platform and lectern.

Virtual classroom

Scene 5 from Figure 4: the Game Corner, with interactive programming practice stations and memory-card activities.

Game-based practice

Figures 2 and 4. The MetaClassroom environment, with virtual teaching and game-based practice shown below.

MetaClassroom connects immersive programming activities with three-dimensional learning progression. A quasi-experimental comparison with traditional classrooms reported improvements in programming achievement, self-regulated learning, learning beliefs and motivation. The study examines how a designed learning environment can connect teaching, learning and assessment.

Study design. Quasi-experimental comparison of an immersive programming environment and traditional classrooms.

Read summary, abstract and citation

Humanities and Social Sciences Communications 2024

What factors will affect the effectiveness of using ChatGPT to solve programming problems? A quasi-experimental study

Original Figure 1 from the paper: the complete five-stage experimental design, retaining all process descriptions, durations, arrows and original formatting.

Experimental design

The complete example chart from the task sheet in Figure 2, with its axes, legend and values retained; surrounding task instructions are omitted.

Task example

The first complete student output chart from Figure 3, including its title, axes, legend and plotted values.

Student output

Figures 1, 2 and 3. The original experimental design from Figure 1, with the task example from Figure 2 and one student output from Figure 3.

Using Python graphing tasks, this quasi-experimental study examines why learners differ in their effectiveness when using ChatGPT to solve programming problems. AI literacy, programming knowledge and understanding of ChatGPT are associated with effective use. The study connects the design of AI-supported programming activities with learners' existing capabilities.

Study design. Quasi-experimental study in a Python graphing and programming problem-solving context.

Read summary, abstract and citation

Education and Information Technologies 2025

Impact of AI-agent-supported collaborative learning on the learning outcomes of University programming courses

The collaborative programming workflow and its connections to AI agents, large language models, semantic memory and episodic memory.
Figure 1. AI-agent-supported collaborative programming approach.

This six-week study compares AI-agent-supported collaborative learning with computer-supported collaborative learning in a university programming context. The 45 participants were allocated to groups of 24 and 21. The AI-supported group achieved stronger learning outcomes, self-efficacy and interest, with lower mental effort; the difference in mental load was not significant.

Study design. Six-week quasi-experimental comparison, with 45 undergraduates in an ACM programming competition teaching environment.

Read summary, abstract and citation

International Journal of Human–Computer Interaction 2026

Human-AI Collaborative Learning Ecosystem: Effects of Multi-Agent-Based Programming Learning on Learning Outcomes, Computational Thinking, and Behavioral Patterns in Higher Education

A four-part programming instructional framework for passive, active, constructive and interactive learning supported by multiple AI agents.
Figure 1. MA-PL supported intelligent programming instructional framework.

A 12-week experiment with 54 undergraduates compares Multi-Agent-based Programming Learning (MA-PL) with conventional programming instruction. Students receiving multi-agent support achieved stronger learning outcomes and computational thinking, with higher-level cognitive engagement in their programming behavior. Behavioral analysis also identified less autonomous construction and a shift from peer collaboration toward human-AI collaboration.

Study design. 12-week randomized experiment with 54 undergraduates (27 per group), combining learning-outcome and computational-thinking assessments with behavioral sequence analysis.

Read summary, abstract and citation

Generative AI policy and adoption

Educational technology research and development 2025

The journey of challenges and victories: exploring the transformation action framework in the GenAI era from multifaceted policies

Figure 2: general large models, educational data and education-specific models leading to new educational forms.

A Educational transformation

Figure 3: three models of relationships between teachers, students and computers in education.

B Classroom relationships

Figures 2 and 3. From education-specific models to changing classroom relationships.

This study synthesizes 11 policy and guideline documents through meta-ethnography. Its framework connects macro-level social and workforce change, meso-level educational organization and pedagogical logic, and micro-level teaching and learning scenarios. It offers a way to examine how generative AI policy relates to changes in educational practice.

Study design. Meta-ethnographic synthesis of 11 policy and guideline documents from international organizations, education ministries and universities.

Read summary, abstract and citation

Journal of Computer Assisted Learning 2025

Deconstructing University Learners' Adoption Intention Towards AIGC Technology: A Mixed-Methods Study Using ChatGPT as an Example

The research model connecting four AI literacy dimensions and planned-behavior factors to configurations of AIGC adoption intention.
Figure 3. Research conceptual model.

This mixed-methods study examines university learners' intentions to adopt AIGC technology using ChatGPT as its example. Exploratory activities and interviews inform a decision framework, followed by fuzzy-set qualitative comparative analysis of 233 valid questionnaires. Six configurations are associated with high adoption intention, with AI literacy and perceived behavioural control playing prominent roles.

Study design. Sequential exploratory and configurational analysis using interviews and fsQCA of 233 valid questionnaires.

Read summary, abstract and citation

Education and Information Technologies 2024

Empowering education development through AIGC: A systematic literature review

A two-path, four-level framework linking the core advantages and potential risks of AIGC to educational application scenarios.
Figure 7. The underlying mechanism and logical framework of AIGC's educational application.

This systematic review examines 134 publications retrieved from EBSCO, EI Compendex, Scopus and Web of Science. It maps educational uses of AIGC across performance assessment, instructional applications, learning outcomes, advantages and disadvantages, and future prospects. Grounded-theory analysis informs an account of application scenarios and a research agenda.

Study design. Systematic literature review of 134 publications from four databases, supplemented by grounded-theory analysis.

Read summary, abstract and citation

The International Journal of Management Education 2026

To use or not to use? Generative AI adoption in Chinese business schools

Estimated paths from positive and negative valence factors to generative AI adoption intention and use, including the moderating role of institutional policy.
Figure 2. Results of path analysis: the valence framework.*p < 0.05; **p < 0.01; ***p < 0.001.

A survey of 797 students and faculty applies the valence framework to GenAI adoption in Chinese business schools. PLS-SEM links expected benefits, enjoyment and AI literacy with stronger adoption intentions, and anxiety, hallucination and privacy concerns with weaker intentions. Institutional policy moderates the intention-use relationship, with different adoption patterns for students and faculty.

Study design. Survey of 797 students and faculty, analyzed using PLS-SEM, institutional-policy moderation and student-faculty multigroup comparisons.

Read summary, abstract and citation

The International Journal of Management Education 2026

From information evaluation to technology use: Generative AI adoption in Chinese business schools

Estimated paths connecting argument quality, source credibility and perceived ease of use with usefulness, intention and actual generative AI use, including expertise as a moderator.
Figure 2. Results of path analysis: information evaluation and technology use.*p < 0.05; **p < 0.01; ***p < 0.001.

Integrating the Information Adoption Model and Technology Acceptance Model, this survey of 672 business school students examines GenAI adoption. Ease of use, argument quality and source credibility are associated with perceived usefulness, which relates to adoption intention and reported use. Expertise and demographic and experience differences shape these relationships.

Study design. Survey of 672 business school students, analyzed using PLS-SEM, expertise moderation and demographic and experience-based multigroup comparisons.

Read summary, abstract and citation

Self-regulated learning and learning analytics

British Journal of Educational Technology 2026

AI support in self-regulated learning: A decade of technological evolution and meta-analysis

Figure 3: the distribution of 133 effect sizes by AI paradigm and publication period, retaining the original axes and legend.

A Evolution of the evidence base

Figure 4: the conceptual intersection of AI evolution and the forethought, performance and self-reflection phases of self-regulated learning.

B AI support across learning phases

Figures 3 and 4. The evidence base across AI paradigms and the self-regulated learning framework.

A meta-analysis of 35 empirical studies from 2013 to 2025 examines AI support within Zimmerman's cyclical model of self-regulated learning. It reports a moderate overall effect (g = 0.507), with different effects across learning phases and measurement approaches. The synthesis distinguishes immediate task performance from durable, transferable self-regulatory competence.

Study design. Meta-analysis of 35 empirical studies, with SRL-phase, technology and contextual moderator analyses.

Read summary, abstract and citation

Data in Brief 2024

Annotated emotional image datasets of Chinese university students in real classrooms for deep learning

Figure 1: classroom recording and facial-image extraction, retaining the anonymization in the published paper.

A Classroom data collection

Table 2, re-typeset: Surprise 218 images; Neutral 1611; Happy 592; Confuse 1176; Boring 1930; Total 5527. Original category labels and values are unchanged.

B Data distribution

Figure 1 and Table 2. Data collection and the composition of the classroom image dataset. Table 2 is re-typeset with its original labels and values.

This data article describes annotated emotional images of Chinese university students in real classroom settings for deep-learning research. It provides a reference point for studying classroom affect and evaluating image-based educational analytics using the dataset's documented collection and annotation procedures.

Study design. Data article describing annotated classroom emotional images for deep learning.

Read summary, abstract and citation

Educational agents and simulation

Interactive Learning Environments 2026

GenAI-agents in education: a systematic review of macro-level trends, educational roles, theoretical foundations and future agendas

Figure 4: four educational GenAI-agent roles and their connections to literacy development.

A Educational roles

Figure 6: the future agenda from trust and control to contextual adaptation, multi-agent collaboration and ecological balance.

B Future directions

Figures 4 and 6. Educational roles and the future directions of GenAI agents.

This systematic review organizes research on generative-AI agents in education around macro-level trends, educational roles, theoretical foundations and future agendas. It provides a literature-synthesis entry point for examining how educational agents are conceptualized and studied across learning settings.

Study design. Systematic review of GenAI agents in education.

Read summary, abstract and citation

Findings of the Association for Computational Linguistics: EMNLP 2025 2025

Evolution in Simulation: AI-Agent School with Dual Memory for High-Fidelity Educational Dynamics

The simulated AI-Agent School environment, showing classrooms, laboratories, a library, a playground and other school spaces.
Figure 1. Structural diagram of AI-Agent School (AAS).

AI-Agent School models educational interactions through an experience-reflection-optimization cycle and a dual memory architecture. Its experience and knowledge bases combine short- and long-term memory. The work examines how agents evolve through simulated school interactions, providing a computational approach to studying educational dynamics.

Study design. Computational educational simulation using LLM agents, a Zero-Exp strategy and dual memory.

Read summary, abstract and citation