Specializing in AI for Education, Immersive Education, and Extended Reality.
PhD Candidate in HCI, The Hong Kong Polytechnic University
Google Scholar: 216 citations · h-index 7 (July 2026)
Not a paper list, but a set of learning experiences I am building and testing.
I am a PhD candidate at The Hong Kong Polytechnic University's Department of Computing, specializing in Human-Computer Interaction, including Virtual Reality, Metaverse Technology, Education Technology, and Extended Reality.
I am supervised by Prof. Richard Chen Li and co-supervised by Prof. Qing Li.
My research has been published in leading academic journals and presented at top-tier international conferences, including the International Journal of Human-Computer Studies (IJHCS), IEEE Transactions on Learning Technologies (TLT), and the IEEE Conference on Virtual Reality and 3D User Interfaces (IEEE VR).
ICBL 2026
ICTE 2025
ICBL 2025
ICWL 2024
A 29-person scripted-dialogue experiment compared urgency and incentive prompts across healthcare, finance, shopping, and home security. Scenario differences matter — but self-report scores are not real disclosure, and questionnaire effect strength cannot be read directly as trust direction.
In two U.S. survey waves, the share of men and women who had ever used an LLM converged, yet frequency of use and self-rated proficiency still differed. Using more often is not the same as being more skilled — and not using is not necessarily falling behind. After diffusion, ask which tasks people keep the tool in.
After 62 people used the app for a week, scores rose for sexual-health knowledge, confidence, and risk-avoidance intention — yet information credibility and chat privacy remain separate concerns. STI Shield splits fixed lessons from open dialogue, connecting hard-to-ask questions with step-by-step learning needs.
In a 48-person trial, the group that generated their own VR scenes scored higher on mindfulness but showed no extra anxiety relief. Some treated prompting like gardening; others worried about performing poorly. When resting, when does authorship become a burden?
In a controlled study of 20 participants, the Agent condition yielded higher task correctness and less estimated human effort time, yet future-use preference still leaned toward Copilot. After automation executes, understanding and control still need a person in charge.
A CHI 2026 survey of 85 studies extends design-by-analogy from inspiration into prototyping, fabrication, and evaluation. What matters is what you borrow, how you borrow it, and where users can still redirect the process.
