Machine learning researcher at Tencent working on LLM reasoning, long-form generation, and agent context management
3
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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—
Output scoring is unavailable: the underlying data does not yet include a verified full publication count or venue breakdown for this author, so this axis, which measures the volume and consistency of research output over time, cannot be scored.
Impact—
Impact cannot be scored yet: citation-based signals are not sufficient (h-index 1 and i10-index 0 leave no stable impact signal), and this axis would measure citation uptake and influence on subsequent research.
Focus—
Focus is not scored: although topic labels exist in the snapshot, the data needed for a field-concentration measure is incomplete; this axis would measure how narrowly or broadly the research agenda is concentrated.
Collaboration53/100
Collaboration scores 53/100, a mid-range value. The OpenAlex snapshot records 13 distinct coauthors across 17 works in the last three years, and the repeated-coauthor list includes Zhuoshi Pan, Junru Lu, H. Vicky Zhao, Xing Sun, Junjie Huang, Jiarui Qin, Weiwen Liu, Yong Yu, Weinan Zhang, Qizhi Pei, and Honglin Lin. The pattern suggests steady teamwork within a bounded group at Tencent and partner institutions rather than an extensive external network.
Activity84/100
Activity scores 84/100, the strongest signal in this profile. Counts by year rise from 2 works in 2024 and 1 in 2025 to 14 in 2026, and the snapshot totals 17 works over the trailing three years — a steep ramp concentrated in the current year. All three Evolve-catalog papers are 2026 arXiv preprints, consistent with an intense recent release cadence whose citation effects have not yet had time to appear.
Reproducibility—
Reproducibility is not scored: none of the reviewed papers carry an Evolve repro-readiness assessment, so this axis, which measures whether code, data, and protocols support independent replication, has no basis for a score.
Research summary
Di Yin is a machine learning researcher at Tencent (China) whose recent work centers on large language models: how they reason, how they use evidence in long-form generation, and how agents manage context. The bibliometric snapshot, name-resolved from OpenAlex, records 17 works over the trailing three years — 2 in 2024, 1 in 2025, and 14 in 2026 — an activity profile that scores 84/100, against a mid-range collaboration score of 53 built on 13 distinct coauthors. Cumulative citation-based indicators remain small (h-index 1, i10-index 0), typical for a record whose output is concentrated in the current year and has had little time to accrue citations. Top-assigned topics are Topic Modeling (10), Multimodal Machine Learning Applications (7), Natural Language Processing Techniques (6), AI-based Problem Solving and Planning (5), and Information Retrieval and Search Behavior (4), sketching an NLP/AI generalist rather than a narrow specialist. Within the Evolve-reviewed catalog, Di Yin appears as a middle author on three 2026 arXiv papers: retrospective process supervision for evidence-grounded long-form generation; a probe of LLM epistemic myopia under long-tail divergent knowledge; and ContextPilot, fine-grained RL for proactive context management in agents. No peer-review venue, field labels, or reproducibility assessment is available for these papers yet, so this should be read as an early-stage, preprint-centric record.
Collaboration pattern
Di Yin's collaboration graph is compact: 13 distinct coauthors over 17 recent works yield a mid-range collaboration score of 53/100. Recurring partners include Zhuoshi Pan, Junru Lu, H. Vicky Zhao, Xing Sun, Junjie Huang, Jiarui Qin, Weiwen Liu, Yong Yu, Weinan Zhang, Qizhi Pei, and Honglin Lin — a mix consistent with a Tencent core group plus academic collaborators. All catalog papers are multi-authored, with Di Yin in a middle rather than lead position.
Research keywords
Highlights
From final artifacts to trajectories: retrospective process supervision
2026 arXiv preprint with Di Yin as a middle coauthor; proposes supervising the generation process rather than only final outputs, targeting evidence-grounded long-form generation. No venue or repro-readiness data is recorded in the catalog.
Probing LLM epistemic myopia under long-tail divergent knowledge
2026 arXiv preprint ('Blind Men and the Elephant') with Di Yin as a middle coauthor; examines how LLMs behave when faced with long-tail, divergent knowledge. Field labels and reproducibility fields are empty.
ContextPilot: proactive context management via fine-grained RL
2026 arXiv preprint with Di Yin as a middle coauthor; trains agents to manage context proactively using fine-grained reinforcement learning. The catalog entry carries no venue or repro-readiness assessment.
Timeline
- 2024First indexed works appear, with 2 publications recorded in the OpenAlex counts-by-year data.
- 2025A quiet year with only 1 indexed work, before the sharp 2026 ramp-up.
- 2026Output peaks at 14 works, including all three Evolve-catalog arXiv papers on process supervision, LLM epistemic myopia, and ContextPilot.
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.
- Bibliometric indicators (h-index, i10-index, counts by year, topics) come from an OpenAlex name-resolved snapshot; Di Yin is a common Chinese name, so identity resolution should be treated with care.
- Only three Evolve-reviewed papers (all 2026 arXiv preprints) are available; no peer-reviewed venue, field labels, or reproducibility assessments exist for them.
- Citation-based impact is not yet meaningful: h-index 1 and i10-index 0 reflect a young record rather than measured influence.
- Radar scores for output, impact, focus, and reproducibility are missing due to insufficient underlying data.