Evaluation-first AI researcher probing LLM graph agents, evidence provenance, and legal document understanding (early-stage profile)
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-28) 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—
Not yet scored: no publication counts, venue history or per-year productivity metrics are available in this snapshot. This axis would measure total scholarly output and how it is distributed over time.
Impact—
Not yet scored: citation counts, h-index and downstream-adoption indicators are missing from the current data. This axis would measure how widely this researcher's papers are cited and built upon by others.
Focus100/100
The radar assigns focus a maximum score of 100: the raw field distribution contains exactly one field, artificial intelligence (LLM agents / retrieval-augmented generation), and the reviewed GraphEcho paper sits squarely in controlled benchmark evaluation. Both cataloged 2026 preprints — graph-agent provenance and legal document understanding — share an evaluation-first agenda, indicating a tightly concentrated research identity, though the score rests on only two papers.
Collaboration56/100
Scoring 56, this axis rests on 16 distinct coauthors across the cataloged work. The frequent-coauthor list repeats Lei Fan and Fan Zhang, alongside Sikun Wang, Yankai Chen, Zhuohan Xie, Sijia Peng, Xinhua Ji, Cunyuan Zheng, Huangyong Shan and Philip S. Yu. The pattern suggests moderately broad but repeat-anchored teamwork, with Zhou in a middle author position on both papers rather than leading as corresponding author.
Activity—
Not yet scored: recent-activity signals such as publication cadence and preprint posting rhythm are absent beyond the two 2026 entries. This axis would track ongoing research engagement in recent years.
Reproducibility55/100
Reproducibility scores 55, a mean (reproMean 55) taken over just one reviewed paper (reproPapers 1): GraphEcho. Evolve's verdict credits rigorous statistics but flags a single training seed and a narrow set of models, with reproducibility readiness rated medium. With a sample of one, this number describes that benchmark's artifact practices rather than a stable personal track record.
Research summary
Yixi Zhou is an AI researcher whose Evolve footprint is early-stage but sharply focused: two 2026 arXiv preprints make up the catalog to date. 'GraphEcho: Structural Redundancy and Evidence Provenance in LLM Graph Agents' builds a controlled benchmark separating path redundancy from evidence provenance in LLM graph agents, using GRPO-based reinforcement learning and source-counting probes; Evolve's review credits it with rigorous statistics while cautioning that a single training seed and a narrow model pool temper the conclusions, and rates its reproducibility readiness as medium. The second preprint, 'Same Scores, Different Decisions: Evaluating JEV and Language Models for Legal Document Understanding', extends the same evaluation-first instinct to legal documents, asking whether identical scores conceal different decisions. The radar assigns a perfect focus score of 100 — the raw field distribution contains exactly one field, artificial intelligence (LLM agents / retrieval-augmented generation) — plus a collaboration score of 56 across 16 distinct coauthors, and a reproducibility score of 55 computed from a single reviewed paper. Output, impact and activity axes remain unscored because their underlying metrics are missing. Zhou appears in a middle author position on both papers, without corresponding-author roles, within a network anchored by repeat collaborators.
Collaboration pattern
Yixi Zhou appears as a middle author on both cataloged 2026 preprints, neither as corresponding author. The collaboration score is 56 over 16 distinct coauthors, with Lei Fan and Fan Zhang recurring in the frequent-coauthor list alongside Sikun Wang, Yankai Chen, Zhuohan Xie, Sijia Peng, Xinhua Ji, Cunyuan Zheng, Huangyong Shan and Philip S. Yu — a moderately sized network anchored by repeat collaborations rather than a single long-term pair.
Research keywords
Highlights
Controlled benchmark separates path redundancy from evidence provenance in LLM graph agents
Same scores, different decisions: probing JEV and language models on legal documents
A second 2026 arXiv preprint carries the evaluation-first approach into legal document understanding, examining whether JEV and language models that earn identical scores actually support the same downstream decisions — a study of what automatic metrics hide.
Timeline
- 2026Co-authored the arXiv preprint 'Same Scores, Different Decisions: Evaluating JEV and Language Models for Legal Document Understanding', in a middle author position.
- 2026Co-authored the arXiv preprint 'GraphEcho: Structural Redundancy and Evidence Provenance in LLM Graph Agents', which becomes the first Evolve-reviewed paper in the catalog, with a controlled-benchmark verdict and medium reproducibility readiness.
- 2026As of September 2026, the profile stands at two arXiv preprints, one classified field (LLM agents / retrieval-augmented generation), 16 distinct coauthors, and reproducibility evidence from a single reviewed paper.
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.
- Radar scores are computed from public bibliometric data; output, impact and activity metrics were missing from this snapshot, so those axes are unscored rather than zero.
- No verified institution, country or ORCID is attached; the profile is name-resolved from public bibliographic data and may conflate researchers sharing the name Yixi Zhou.
- Focus (100) and reproducibility (55) rest on a very small sample — one reviewed paper and two cataloged preprints — and may shift substantially as more work is reviewed.
- Both cataloged papers are 2026 arXiv preprints; peer-reviewed venue acceptance is not recorded in the data.