Research Profile
Social recommendation / Causal reasoning / Graph learning
Building recommender systems that understand social behavior, not only social links.
I study how people form preferences, friendships, and social influence patterns in online platforms. My recent work combines graph neural networks, causal inference, dynamic graph learning, and large language models to make recommender systems more accurate, more interpretable, and more faithful to real social mechanisms.
Research Focus
Social Recommendation
Modeling when social ties truly help recommendation, when they introduce noise, and how recommender systems can distinguish preference similarity from social influence.
Causal Reasoning
Using counterfactual thinking to refine social graphs, estimate friend influence, and avoid treating every observed social link as equally useful evidence.
Graph and LLM Methods
Designing graph learning models that combine structural signals with textual, personality, temporal, and diffusion features from real social platforms.
Selected Work
Counterfactual refinement for graph-based social recommendation
A causal approach for identifying whether friends actually influence user choices and using that signal to refine noisy social graphs.
Feature-driven dynamic graph learning for cascade popularity prediction
A dynamic graph model that combines microscopic node-level diffusion patterns with macroscopic cascade-level features.
LLM-enhanced friend recommendation through personality disentanglement
A model that uses LLM-derived personality traits to partition social graphs and explain different paths to friendship formation.