Ph.D. Student at Tongji University · Visiting Student at CUHK
I am interested in structured and generative intelligence, with current work on knowledge graph completion (triple set prediction), generative recommendation, and agent memory.
I am a Ph.D. student at Tongji University, where I also completed my undergraduate study.
I am advised by Prof. Jihong Guan and Prof. Wengen Li.
Since 2026, I have been visiting The Chinese University of Hong Kong (CUHK), advised by Prof. Hong Cheng.
News
Selected updates
2026.08
Started a research visit at CUHK with Prof. Hong Cheng.
2026
Released DiffTSP, a discrete diffusion framework for knowledge graph triple set prediction.
2026
OKG-LLM was published in TKDE.
2025
Raker was published in TKDD.
Education & Experience
2026 – Present
The Chinese University of Hong Kong
Visiting Student · Host: Prof. Hong Cheng
Ph.D.
Tongji University
Ph.D. · Advisors: Prof. Jihong Guan and Prof. Wengen Li
B.Eng.
Tongji University
Publications
Adaptive Item-based Collaborative Structures via Noise Rescheduling in Diffusion for Generative Recommendation
My note.
This work introduces item-based collaborative structure into discrete diffusion recommendation,
using collaborative Semantic IDs and adaptive noise rescheduling to make denoising more aware of item-level dependencies.
One Pass for All: A Discrete Diffusion Model for Knowledge Graph Triple Set Prediction
My note.
This work moves knowledge graph completion from query-level link prediction toward set-level generation, modeling missing triples jointly through discrete diffusion.
OKG-LLM: Aligning Ocean Knowledge Graph with Observation Data via LLMs for Global Sea Surface Temperature Prediction
My note.
This work explores how structured ocean knowledge can be aligned with numerical observations and incorporated into LLM-based time-series forecasting.
Raker: A Relation-Aware Knowledge Reasoning Model for Inductive Relation Prediction
My note.
This work studies inductive relation prediction when useful paths are absent or incomplete, and explores relation-aware neighboring evidence as an alternative reasoning signal.
Research
Broadly, I am interested in structured prediction and generation:
how models can reason over relations, generate coherent sets of outputs,
and retain useful information over long-term interaction.
