Publications & Manuscripts

* Equal contribution. Work under review is labeled below.

GUIHub framework: adapter profiling builds capability vectors, target-level adaptation selects LoRA adapters, and request-time serving reuses their weighted composition for GUI actions.

GUIHub: Privacy-Preserving GUI Agent Collaboration via Capability-Driven LoRA Composition

Xuerong Liu*, Yike Jin*, Hongming Piao*, Shuguang Wang, Dapeng Wu, Ying Wei

Under review · 2026

Capability-driven collaboration among GUI agents through application-specific LoRA adapters, capability-aware routing, and expert composition. My work includes training 14 adapters and evaluating composition and retrieval baselines across IID, OOD, evolving-hub, and storage-constrained settings.

MethodMate workflow: proactive conversation guidance, paper search and review, quantitative research proposal generation, evaluation, iteration, and finalization through user–LLM collaboration.

MethodMate: Supporting Novice HCI Researchers in Designing Quantitative Research Proposals with LLMs

Zejian Li, Qi Liu, Yike Jin, Yixiang Zhou, Rongjie Mao, Pei Chen

Under review · 2026

A literature-grounded LLM system that supports quantitative research proposal design. I developed the full-stack system and contributed to formative research, workflow design, and three user studies.

LLM sycophancy study procedure: pre-task surveys, investment and speed-dating decisions with randomly assigned AI feedback conditions, and post-task workload and trust surveys.

Does Sycophancy Change Decisions? Effect of LLM Sycophancy on AI-Assisted Decision-Making

Zejian Li, Jiaman Pan, Qi Liu, Yuning Xi, Yixiang Zhou, Yike Jin, Rongjie Mao, Pei Chen

Paper ↗

CHI 2026

Investigating how opinion agreement, direct praise, and self-deprecation affect AI-assisted decisions. I contributed to the experimental platform and a mixed-design online study with 106 participants.