LLM evaluation researcher: agent benchmarks, diverse solution search, and large-scale model architecture
3
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
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 data is not available in this snapshot: total-work and per-year publication counts are too sparsely populated to score. This axis would measure the scale and year-over-year consistency of the publication record.
Impact—
Impact metrics are not reliably populated: the snapshot records h-index and i10-index as zero, so this axis is left unscored. It would measure citation uptake and field-level influence of the work.
Focus—
Focus scoring is unavailable: only five coarse topic tags exist (two works each for topic modeling and NLP techniques), too few to compute concentration. This axis would measure how concentrated the research agenda is.
Collaboration80/100
Scores 80/100. The snapshot records 53 distinct coauthors, indicating a broad collaboration network, and Evolve's catalog lists frequent collaborators including Wei Fan, Xinjie Shen, Xudong Guo, Jianhong Tu, Yang Su, and Yangqiu Song. All three cataloged papers sit in the last author position, a pattern consistent with senior or coordinating contributions to team-produced agent-evaluation and model-architecture work.
Activity40/100
Scores 40/100, based on a thin recent record: recent-3-year works number 3, and the per-year count shows 3 works in 2026 with nothing recorded in earlier years. All three cataloged papers appeared in 2026, so measured activity concentrates in one year; the low score may reflect incomplete index coverage as much as low output.
Reproducibility—
No aggregate reproducibility score exists yet. This axis would measure openness of code, data, and experimental protocols. Evolve rates one reviewed paper (E-Commerce Bench) high on reproducibility readiness, but a single item cannot support a score.
Research summary
Dayiheng Liu is a language-model researcher whose recent cataloged output centers on evaluation and benchmarking of large language model (LLM) systems. Evolve's catalog lists three 2026 arXiv papers, all with Liu in the last author position: E-Commerce Bench, a long-horizon benchmark for LLM agents running autonomous business operations; Answer Probing-Guided Search for Diverse Solution Exploration of LLMs; and a design study of the Qwen3.8-Next architecture covering evaluation, efficiency, and training stability. The OpenAlex topic profile is consistent with this spread — topic modeling (2 tagged works), natural language processing techniques (2), plus single works on advanced neural network applications, multi-agent systems and negotiation, and AI-based problem solving and planning. Bibliometric coverage in the snapshot is thin: it records 3 works in the last three years and 53 distinct coauthors, which yields a collaboration score of 80 but an activity score of only 40, while the citation-based output and impact axes lack sufficient data to score. The one paper Evolve has fully reviewed, E-Commerce Bench, is judged the first deterministic year-long e-commerce agent benchmark, with a strong reproducible open design (repro readiness: high), tempered by 5-run variance and vendor self-evaluation in the rankings.
Collaboration pattern
The snapshot records 53 distinct coauthors and a collaboration score of 80, pointing to a wide, team-based working pattern. Recurring collaborators include Wei Fan, Xinjie Shen, Xudong Guo, Jianhong Tu, Yang Su, Yinger Zhang, Lianghao Deng, Fengyu Wang, Baohua Dong, Yangqiu Song, Zihan Qiu, and Zekun Wang. All three cataloged papers place Liu in the last author position, consistent with a convening or supervisory role on large model-building and evaluation efforts.
Research keywords
Highlights
First deterministic year-long e-commerce agent benchmark
Last-author study of the Qwen3.8-Next architecture
The catalog lists a 2026 arXiv paper, On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability, with this researcher in the last author position; Evolve has not yet published a verdict, keywords, or repro-readiness rating for it.
Answer probing-guided search for diverse LLM solutions
The catalog also lists the 2026 arXiv paper Answer Probing-Guided Search for Diverse Solution Exploration of LLMs, again with the researcher as last author; no Evolve review verdict or keywords are recorded yet.
Timeline
- 2026Snapshot records 3 works in 2026 (recent-3-year works: 3), with all cataloged papers falling in this year.
- 2026Last-author arXiv paper proposing E-Commerce Bench, a deterministic year-long benchmark for LLM agents in autonomous e-commerce operation.
- 2026Last-author arXiv paper on answer probing-guided search for diverse solution exploration of LLMs.
- 2026Last-author arXiv paper on the design of the Qwen3.8-Next architecture, covering evaluation, efficiency, and training stability.
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 snapshot comes from OpenAlex only; citation counts, h-index, and i10-index appear unpopulated (recorded as zero) and are therefore not interpreted.
- No institution, country, or ORCID is linked; the profile is name-resolved from public bibliographic data and identity merging or splitting cannot be ruled out.
- Only 3 papers are in Evolve's catalog; two of them lack a review verdict, keywords, and repro-readiness assessment.
- Work coverage is limited to the snapshot window (3 works in the recent 3 years); earlier publications are not represented.