Yufan Zhao

Yufan Zhao

Paper Library
Research Plan

LLM & Synthetic Data for RecSys

57 papers from 2025–2026 on cleaning interaction logs, cross-domain synthetic events, grounded augmentation, recommendation curricula, and user simulation.

Reasoning over Semantic IDs

71 papers from 2025–2026 on when intermediate computation helps item-token prediction: prefix rewards, latent reasoning, explicit rationales, retrieval, and test-time search.

LLM × User Profiles

69 papers from 2025–2026 on writing profiles from behavior, updating them over time, grounding claims in evidence, and testing whether models actually use them.

User Interest Exploration

45 papers from 2025–2026 on testing whether a new interest is real, proposing nearby directions, giving users control, and measuring long-term effects.

GRPO after the Boom

80 papers from 2025–2026 on GRPO objectives, failure modes, entropy and diversity, rollout efficiency, domain adaptations, and simpler alternatives.

Technical Reports Worth Reading

92 reports from 2025–2026 covering Qwen, DeepSeek, Kimi, GLM, Gemma, Nemotron, multimodal branches, agents, and industrial recommendation.

Semantic IDs & Alternatives

90 papers from 2025–2026 testing where Semantic IDs fail and when text, summaries, generated vectors, soft or hybrid representations, and ID-free interfaces work better.
© 2026 Yufan Zhao · yufzhao-studio.org