Researcher advancing generative recommendation with semantic IDs via reinforcement learning and credit assignment
2
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-27) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Not yet scored: this profile lacks a resolved, complete publication history, so output volume cannot be computed. This axis would measure the number of papers over time and the shape of the publishing trajectory.
Impact—
Not yet scored: no citation-level data is available for this name. This axis would measure citation counts and downstream uptake of the reviewed papers, which cannot be inferred from two recent preprints.
Focus—
Not yet scored: topical concentration cannot be estimated from only two reviewed 2026 preprints. This axis would measure how research effort concentrates across fields and topics over a longer publication record.
Collaboration41/100
Evolve scores collaboration at 41 on a 0-100 scale, computed from real bibliometric data. The raw record shows 7 distinct coauthors, and the recurring list is stable: Mengdan Zhu, Yufan Zhao, Yao Zhao, Tao Di, Yulan Yan, Sridhar Iyer, and Liang Zhao. Both catalog papers are middle-authorship team efforts, consistent with a compact, concentrated working group rather than a broad, high-churn collaboration network.
Activity—
Not yet scored: no year-by-year publication record is available beyond the two 2026 papers. This axis would measure recent publishing cadence and sustained research activity across calendar years.
Reproducibility—
Not yet scored at profile level. This axis would measure code, data, and reporting practices across a body of work; individually, both reviewed papers carry a medium reproducibility-readiness tag, which is not a profile-level score.
Research summary
Sophie Di is a researcher working at the intersection of generative recommendation and reinforcement learning; this profile is name-resolved from public bibliographic data. The Evolve catalog holds two 2026 arXiv preprints, both in recommender systems and information retrieval, both with middle-position authorship and no corresponding-author role. The first, From Interests to Semantic IDs, introduces retrieval-grounded credit assignment for semantic-ID generative recommenders, built on group-relative policy optimization and process supervision; Evolve's review credits it with improvements across three datasets and good ablations, while flagging partial significance and untested reward leakage. The second, Learning Better Reasoning for Generative Recommendation with Semantic IDs, presents Evo-Rec, which pairs best-of-n chain-of-thought SFT with GRPO under an NDCG-based reward and lifts performance on three Amazon datasets; Evolve notes solid ablations but no seed variance and second-hand baselines. Both carry a medium reproducibility-readiness tag, and together they define a coherent agenda: making semantic-ID generative recommenders trainable through reward design, credit assignment, and process-level supervision. The collaboration footprint is compact — 7 distinct coauthors, a stable recurring group including Mengdan Zhu, Yufan Zhao, Yao Zhao, Tao Di, Yulan Yan, Sridhar Iyer, and Liang Zhao — yielding a collaboration score of 41 on Evolve's 0-100 scale. Output, impact, focus, activity, and reproducibility dimensions are not yet scored.
Collaboration pattern
The bibliometric snapshot records 7 distinct coauthors, and the recurring list is stable: Mengdan Zhu, Yufan Zhao, Yao Zhao, Tao Di, Yulan Yan, Sridhar Iyer, and Liang Zhao. Both reviewed 2026 papers are middle-authorship team efforts without a corresponding-author role, consistent with contributing inside a fixed group rather than leading solo lines. Evolve's collaboration score of 41 (0-100) reflects this compact, concentrated network.
Research keywords
Highlights
Retrieval-grounded credit assignment for semantic-ID recommenders
Evo-Rec: reasoning SFT with NDCG-reward GRPO
A coherent RL reward-design agenda for generative recommendation
Across both reviewed 2026 preprints the same recipe recurs: semantic IDs plus GRPO-style reinforcement learning with carefully shaped rewards (process supervision, retrieval grounding, NDCG-based reward), each evaluated on three datasets — a focused, method-driven line with consistent medium reproducibility readiness.
Timeline
- 2026First cataloged preprint: From Interests to Semantic IDs on retrieval-grounded credit assignment (arXiv, middle authorship).
- 2026Second cataloged preprint: Evo-Rec applies best-of-n reasoning SFT with NDCG-reward GRPO (arXiv, middle authorship).
- 2026Bibliometric snapshot records 7 distinct coauthors and a collaboration score of 41, with a stable recurring coauthor group.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Identity is name-resolved from public bibliographic data; no ORCID, institution, or country is attached, so affiliation and disambiguation should be verified.
- Bibliometrics (collaboration score of 41, 7 distinct coauthors) come from Evolve's OpenAlex-derived snapshot; citation counts and h-index are not available.
- Coverage is limited to the two reviewed 2026 arXiv preprints; other publications by this author may exist outside the Evolve catalog.
- Both papers are arXiv preprints; peer-review status and final publication venues are not reflected in this profile.