Tao Di

—

Unclaimed · AI-structured profileOpenAlex

Recommender systems researcher advancing generative recommendation with semantic IDs and RL-based credit assignment

papers here

reviewed by Evolve

first-author

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—

No output score yet: the snapshot lacks a reliable total-publication count attributable to this identity. This axis would measure cumulative publication volume, venue distribution, and how the output trajectory develops over time.

Impact—

No impact score yet: citation evidence for this identity is too thin to benchmark (h-index 4, i10-index 2 are reported but very small). This axis would measure citation-based influence, including aggregate citations and field-normalized impact.

Focus—

No focus score yet: topical labels are scattered across recommender systems, graph neural networks, topic modeling, renewable energy, and cyclone separators, blocking a coherent concentration measure. This axis would measure topical concentration of the research agenda.

Collaboration41/100

Collaboration scores 41 out of 100. The OpenAlex record lists 7 distinct coauthors, and the Evolve catalog shows a recurring group: Mengdan Zhu, Yufan Zhao, Sophie Di, Yao Zhao, Yulan Yan, Sridhar Iyer, and Liang Zhao. Both reviewed 2026 papers place Tao Di in a middle author position without corresponding status, consistent with specialist-contributor roles inside small teams rather than solo or senior-lead authorship.

Activity52/100

Activity scores 52 out of 100. The snapshot reports 5 works in the last three years, and year-by-year counts show a steady recent cadence: one to two works in most years from 2019 through 2024, a peak of 3 in 2025, and 1 in 2026. Sparse earlier entries (2005, 2016) may reflect indexing gaps or name ambiguity rather than a continuous publishing history.

Reproducibility—

No reproducibility score yet: code and data release metrics are absent from the snapshot. This axis would measure artifact availability and verification readiness; Evolve's qualitative review tagged both 2026 papers as medium repro-readiness.

Research summary

Tao Di is a recommender-systems researcher whose recent work centers on generative recommendation built on semantic IDs. This profile is name-resolved from public OpenAlex bibliographic data; no institutional affiliation or ORCID was available, so identity confidence is limited. The Evolve catalog reviews two 2026 arXiv preprints, both with middle author positions and no corresponding role: 'Learning Better Reasoning for Generative Recommendation with Semantic IDs' (Evo-Rec, combining best-of-n rejection-sampling SFT with NDCG-reward GRPO) and 'From Interests to Semantic IDs' (retrieval-grounded credit assignment for semantic-ID recommendation). Evolve judged both technically solid on ablations while flagging missing seed variance, second-hand baselines, and untested reward leakage. The OpenAlex snapshot reports an h-index of 4, an i10-index of 2, five works in the last three years, and seven distinct coauthors. Topic assignments concentrate on recommender systems (5 works), advanced graph neural networks (4), and topic modeling (3), with thinner clusters in hybrid renewable energy and cyclone separators — a spread that hints at possible homonym conflation in the index. Radar scoring puts activity at 52/100 and collaboration at 41/100, while output, impact, focus, and reproducibility lack sufficient data to score.

Collaboration pattern

The collaboration score is 41/100, based on 7 distinct coauthors in the OpenAlex record. A recurring group — Mengdan Zhu, Yufan Zhao, Sophie Di, Yao Zhao, Yulan Yan, Sridhar Iyer, and Liang Zhao — anchors the network. In both reviewed 2026 papers Tao Di appears in a middle author position without corresponding status, indicating a contributor role within team projects rather than solo or senior-lead authorship.

Research keywords

Generative recommendationSemantic IDsCredit assignmentReinforcement learningGroup-relative policy optimization (GRPO)Chain-of-thought reasoningProcess supervisionRejection samplingGraph neural networksTopic modeling

Highlights

Evo-Rec: reasoning-trained generative recommendation with semantic IDs
Evolve's review of this 2026 arXiv preprint finds that best-of-n rejection-sampling SFT followed by NDCG-reward GRPO lifts semantic-ID generative recommendation across three Amazon datasets, with solid ablations; noted gaps are missing seed variance and second-hand baselines.
Retrieval-grounded credit assignment for semantic-ID recommendation
In this 2026 preprint, credit assignment grounded in the path from user interests to semantic IDs improves SID-based generative recommendation on three datasets; Evolve calls the ablations good while flagging partial significance and untested reward leakage.
Steady recent output with a 2025 peak
The OpenAlex snapshot credits 5 works over the last three years, including a peak of 3 indexed works in 2025, supporting an activity score of 52/100.

Timeline

2005Earliest record in the OpenAlex snapshot: one indexed work, long before the current recommender-systems research line.
2019Indexed output resumes at a low but steady cadence: one work in 2019, then one to two works in most years through 2024.
2021Counts reach 2 works in both 2020 and 2021, the first sustained stretch of the recent publishing phase.
2025Most productive indexed year in the snapshot, with 3 works — the recent peak behind the 52/100 activity score.
2026Two arXiv preprints on generative recommendation with semantic IDs enter the Evolve catalog, both reviewed with middle author positions and no corresponding role.

Papers in the Evolve catalog

Frequent co-authors

Mengdan Zhu×2 Yufan Zhao×2 Sophie Di×2 Yao Zhao×2 Yulan Yan×2 Sridhar Iyer×2 Liang Zhao×2

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.

Bibliometrics (h-index 4, i10-index 2, counts by year, topics) come from a name-resolved OpenAlex profile; no ORCID or institutional affiliation was available to confirm identity.
The topic spread — recommender systems and graph neural networks alongside hybrid renewable energy systems and cyclone separators — suggests possible homonym conflation in the underlying index.
Both reviewed papers are 2026 arXiv preprints with middle author positions and no corresponding-author role; no peer-reviewed venue or full author list was provided.
Output, impact, focus, and reproducibility axes are unscored because the underlying data is insufficient; collaboration (41) and activity (52) are the only radar scores available.