Early-career SJTU researcher building benchmarks for text-to-SQL, LLM reasoning, and industrial multimodal measurement
3
papers here
1
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
2
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—
Publication-volume data is not yet available for this profile: the snapshot lacks a verified total-output count, so no score is shown. This axis would measure cumulative and per-year publication volume across venues and authorship positions.
Impact—
Citation-impact signals are insufficient: the only supplied indicators (h-index 2, i10-index 1) are too sparse to support a meaningful impact score, which would normally aggregate citations, venues, and uptake of this researcher's benchmarks.
Focus100/100
The focus axis scores 100 because the field distribution in the snapshot is maximally concentrated: the only fielded record maps to multimodal machine learning (computer vision). Topic counts reinforce a coherent data-centric cluster — topic modeling (5), spreadsheets and end-user computing (5), software engineering research (4), advanced database systems (4), and data visualization and analytics (4) — feeding benchmark work on text-to-SQL, reasoning, and multimodal measurement.
Collaboration54/100
Collaboration scores 54 on a moderate footing: the snapshot records 14 distinct coauthors over the last three years, with Jianguo Yao, Haibing Guan, and Xijun Li recurring most often. In the reviewed catalog the researcher appears as first author on BIRD-History and middle author on two other 2026 papers, with no corresponding-author role yet — a network that is broad but still maturing in leadership.
Activity79/100
Activity scores 79, reflecting a clear recent surge: 14 works in the last three years, with yearly counts accelerating from 1 (2020) and 3 (2023) to 3 (2025) and 11 in 2026 alone. Three of those 2026 outputs — InSituMeasure, To Think or Not to Think, and BIRD-History — are already in the Evolve review catalog, indicating sustained, current bench work.
Reproducibility20/100
Reproducibility scores 20, the weakest axis, computed from one assessed paper: InSituMeasure carries a low readiness rating, with the review citing thin statistical support and no stated release of benchmark code or data. With only a single paper assessed so far, this score is an early signal rather than a settled record — releasing artifacts would move it quickly.
Research summary
Yunfan Zhou is an early-career researcher at Shanghai Jiao Tong University whose work converges on building rigorous evaluation assets for data- and model-centric AI. The bibliometric snapshot (OpenAlex identity) records an h-index of 2, an i10-index of 1, and 14 works in the last three years, with annual output accelerating sharply to 11 works in 2026. Topic counts sketch a coherent cluster: topic modeling (5), spreadsheets and end-user computing (5), software engineering research (4), advanced database systems and queries (4), and data visualization and analytics (4). Three 2026 papers are in the Evolve catalog. As first author, Zhou leads BIRD-History, a benchmark for history-driven text-to-SQL with fine-grained knowledge annotations, extending the database-and-dialogue thread. InSituMeasure, a middle-author contribution, probes situated measurement grounding in industrial scenes with multimodal large language models: Evolve's review of the 2,922-task real-scene gauge benchmark finds the best of 24 MLLMs reaching only 25.7% joint value-unit accuracy, praising the diagnosis design while flagging thin statistics and no stated artifact release. A third 2026 paper, To Think or Not to Think, tackles allocating reasoning where it helps in large language models. With 14 distinct coauthors over three years — Jianguo Yao, Haibing Guan, and Xijun Li recurring most often — and no corresponding-author paper yet, this is a profile in ascent: benchmark-centric, industrially grounded, and still early.
Collaboration pattern
The snapshot records 14 distinct coauthors over the last three years. Jianguo Yao, Haibing Guan, and Xijun Li each recur twice in the frequent-coauthor list, alongside Zhengdong He, Chao Shen, Xinyuan Li, Zhihai Wang, Qiming Shi, and Yizhou Yang. The reviewed catalog shows one first-author paper (BIRD-History) and two middle-author papers, none as corresponding author — a broad, actively used network in which leadership roles are still emerging.
Research keywords
Highlights
Industrial gauge-reading benchmark exposes MLLM limits
First-authored history-driven Text-to-SQL benchmark
As first author, Zhou contributed BIRD-History (2026), a benchmark for history-driven text-to-SQL with fine-grained knowledge annotations, extending the database-and-dialogue thread in the topic profile (advanced database systems: 4; topic modeling: 5). Evolve has not yet published a verdict or reproducibility rating for it.
Adaptive reasoning allocation for large language models
To Think or Not to Think: Allocating Reasoning Where It Helps (2026) examines when large language models should engage deliberate reasoning. It is listed in the catalog without an Evolve verdict, keywords, or reproducibility rating so far.
Timeline
- 2020Earliest recorded activity: 1 work in the snapshot's yearly counts.
- 20233 works recorded, building an early publication record.
- 20253 works recorded, holding steady output before the breakout year.
- 2026Breakout year: 11 works recorded, including all three catalog papers — first-authored BIRD-History plus InSituMeasure and To Think or Not to Think.
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 figures are computed from an OpenAlex snapshot; the output and impact axes are marked missing because total-publication and citation data are insufficient.
- The reproducibility score rests on a single assessed paper (InSituMeasure); the other two 2026 catalog papers have no verdicts, keywords, or reproducibility ratings yet.
- Identity is name-resolved from public bibliographic data (OpenAlex; ORCID 0009-0009-7814-4390); Yunfan Zhou is a common name, so the record may merge or split homonyms.
- Yearly counts (2020: 1; 2023: 3; 2025: 3; 2026: 11) reflect indexed records only and may lag actual publications; 2026 is still in progress.