Shanghai Jiao Tong researcher spanning NLP and LLM-agent benchmarking, with a steep output ramp into 2026.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
4
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 is not yet scored for this profile: verified publication-volume data needed for normalization is not available. This axis would measure total and first-authored output against field and career-stage baselines.
Impact—
Impact is not yet scored for this profile because citation records are too sparse to evaluate reliably. This axis would measure citation velocity, citing works, and uptake of the benchmark and dialogue outputs.
Focus0/100
The focus axis scores 0/100 because this record splits evenly across two fields — natural language processing and LLM agent evaluation/AI benchmarking each take 0.5 field weight. Topic assignments echo that breadth: Topic Modeling leads with 7 works, then human-technology interaction (5), AI applications (4), NLP techniques (3), and usability/interface design (3). The pattern reads as a deliberate two-front agenda rather than a single concentrated line.
Collaboration63/100
Collaboration scores 63/100 on 23 distinct coauthors in the record. Twelve recur often enough to flag, including Zixuan Wang, Yufan Zhou, Jinzhou Tang, Xinle Yu, Chengjun Wu, Zhaoxiang Feng, and Letian Peng. Both Evolve-cataloged 2026 papers place Ye in a middle author slot rather than first or corresponding, consistent with a contributor role inside larger teams rather than a small stable partnership.
Activity91/100
Activity is the strongest axis at 91/100. Yearly counts climb from 1 work (2021) and 1 (2023) to 3 (2024), 5 (2025), and 14 in 2026, with 22 of the 24 recorded works falling in the last three years. Both cataloged papers landed in 2026. This is an accelerating, almost entirely current publication tempo, carried so far in preprint form.
Reproducibility38/100
Reproducibility averages 38/100 over the 2 papers Evolve has scored. ERPBench rates medium: its paired Solo-versus-Arena design is solid but rests on single runs within one Arena scope. Mind2Dialogue rates low because its gains depend on a closed Oracle teacher model that outside teams cannot easily re-run. Neither cataloged paper currently reaches high readiness.
Research summary
Lyumanshan Ye is a researcher at Shanghai Jiao Tong University whose record, name-resolved from OpenAlex, splits evenly between two fields: natural language processing and LLM agent evaluation and AI benchmarking (0.5 each). The record lists 24 works, 22 of them in the last three years, climbing from 1 work in 2021 to 14 in 2026; h-index 4 and i10-index 2 mark an early-stage citation profile. Topic modeling leads assigned topics (7 works), ahead of human-technology interaction (5) and AI applications (4). Two 2026 arXiv preprints anchor the Evolve catalog. Mind2Dialogue trains human-aware language models from an Oracle that sees simulated user mental states, distilling privileged information into personalization gains that hold across three backbones, though they rest on a closed teacher model. ERPBench evaluates LLM agents on enterprise decision-making across competitive market ecologies, showing rankings flip between Solo and Arena settings (DeepSeek to Gemini). In both, Ye is a middle-listed, non-corresponding author. Distinct coauthors number 23, twelve of them recurring. Mean reproducibility readiness across the two scored papers is 38/100.
Collaboration pattern
Ye works inside sizeable, rotating teams: 23 distinct coauthors appear in the record, with 12 recurring often, including Zixuan Wang, Yufan Zhou, Jinzhou Tang, Xinle Yu, Chengjun Wu, Zhaoxiang Feng, and Letian Peng. In both Evolve-cataloged 2026 papers Ye sits in a middle author position without corresponding status, suggesting a hands-on contributor role within larger NLP and agent-benchmark groups rather than a fixed two-person partnership.
Research keywords
Highlights
Mind2Dialogue: personalization through simulated user minds
ERPBench: agent rankings flip under market competition
Output ramps steeply into 2026
OpenAlex yearly counts rise from 3 works in 2024 and 5 in 2025 to 14 in 2026, with 22 of the record's 24 works falling in the last three years — both Evolve-cataloged preprints land in this 2026 surge.
Timeline
- 2021First indexed work appears in the OpenAlex record (1 work).
- 20231 indexed work, the last quiet year before the output ramp.
- 20243 works indexed; output begins to build.
- 20255 works indexed as the pace continues to climb.
- 2026Peak year: 14 works, including both Evolve-cataloged preprints (Mind2Dialogue, ERPBench).
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.
- All bibliometrics (h-index 4, i10-index 2, yearly counts, coauthor, topic, and field splits) come from an OpenAlex name-resolved record and may be affected by namesake merging or splitting.
- The output and impact radar axes are unscored; only the raw h-index/i10-index and per-year counts are available, without citing-work detail.
- The reproducibility mean (38) rests on only 2 scored papers, both 2026 arXiv preprints.
- In both cataloged papers Ye is a middle-listed, non-corresponding author, so no leadership role is implied.