Recommender Systems
Recommender Systems
I work on recommendation problems where user behavior, retrieval spaces, and deployment constraints change over time.
Signals, retrieval, and prediction
How can recommendation systems model returning users, retrieve efficiently, and remain reliable under industrial-scale constraints?
Scope
User reactivation, generative recommendation, Semantic-ID interface diagnostics, dynamic retrieval, graph embedding, counterfactual watch-time prediction, and efficient CTR modeling form the main technical line.
User reactivation
Recalibrating recommendation when returning users have pre-gap history but no observed behavior during long inactivity gaps.
Generative recommendation
Representing long histories, structured intent, and preference-aware generation.
Dynamic retrieval
Efficient item retrieval and graph embedding under changing user and item behavior.
Code and artifact diagnostics
Inspecting recommendation artifacts such as Semantic-ID interfaces before downstream training, refresh, and deployment.
Trustworthy prediction and efficiency
Counterfactual watch-time prediction and efficient CTR modeling for deployment.