Ying Zhang

—

Unclaimed · AI-structured profileOpenAlex

NLP researcher in document AI and LLM evaluation: page stream segmentation cost and summarization metric saturation

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-13) 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 volume cannot be scored yet: the OpenAlex snapshot lists only 2 works, both from 2026, with h-index and i10-index recorded as 0 — likely an unresolved-identity artifact for this common name. This axis would measure total publication count and productivity over time.

Impact—

Impact is unscored: the available citation indicators (h-index 0, i10-index 0) come from an OpenAlex identity that likely mixes or misses this common name's papers. This axis would measure citation counts, citing works, and influence on follow-up research.

Focus0/100

Focus scores 0 because the record splits evenly across two fields — natural language processing and document AI / page stream segmentation at 0.5 each — and OpenAlex topics fan out over five areas: topic modeling (2), NLP techniques, handwritten text recognition, text classification, and advanced text analysis. The two 2026 papers still share one thread: LLM-based text processing and its evaluation.

Collaboration41/100

Collaboration scores 41 on a modest base: 7 distinct coauthors across the 2 recent works, including Nikhil Reddy Pottanigari, Sepideh Kharaghani, Ramin Fahimi, Noah Bolger, Saverio Vadacchino, and Alejandro Posada. On both 2026 papers Ying Zhang sits in the last author position without a corresponding-author flag, suggesting a senior-supervision role within small, likely early-career teams.

Activity32/100

Activity scores 32: OpenAlex records exactly 2 works in the last three years, both in 2026, with no earlier per-year counts in the snapshot. This thin record makes the current cadence look limited, though two arXiv releases in a single year suggest a concentrated burst rather than sustained long-run publishing under this identity.

Reproducibility55/100

Reproducibility scores 55, averaged over 2 assessed papers, both rated medium readiness. Evolve's verdicts flag a shared weakness: the MSBD segmentation work rests on single-run evidence without significance checks, and the FPS/QPS/EPS scaffold metrics lack validation against human preference — the methods are plausible and checkable, but statistical rigor is the gap.

Research summary

Ying Zhang is an early-stage natural language processing researcher whose visible record centers on two 2026 arXiv papers, both from the last author position, addressing cost and reliability in LLM-based text processing. The first, 'Splitting Documents at Lower Cost,' introduces MSBD, which batches multi-split boundary decisions into a single LLM call for page stream segmentation, holding accuracy near a model-specific knee while cutting inference cost; Evolve rates the idea plausible but flags single-run evidence without significance checks. The second, 'Preserving What Matters,' documents LLM-as-a-judge saturation in summarization evaluation and proposes semantic scaffold metrics — FPS, QPS, and EPS — that preserve score spread, though validity against human preference remains untested. The OpenAlex snapshot splits the field profile evenly (0.5 natural language processing, 0.5 document AI / page stream segmentation) across five topics — topic modeling (2), NLP techniques, handwritten text recognition, text classification, and advanced text analysis — yielding a focus score of 0. Bibliometrics are thin: h-index and i10-index are recorded as 0, 2 works appear in the last three years (activity 32), with 7 distinct coauthors (collaboration 41) and a reproducibility score of 55 across 2 assessed papers, both medium readiness. Given the common name and missing ORCID and institution, this profile is name-resolved from public bibliographic data.

Collaboration pattern

Collaboration is compact: 7 distinct coauthors across 2 recent arXiv works, with recurring names such as Nikhil Reddy Pottanigari, Sepideh Kharaghani, Ramin Fahimi, Noah Bolger, Saverio Vadacchino, and Alejandro Posada. Ying Zhang takes the last author slot on both 2026 papers (non-corresponding), a pattern consistent with supervising or mentoring a small team rather than hands-on first-author experimentation. No institutional affiliation is available, so the organizational base of this network cannot be confirmed.

Research keywords

page stream segmentationdocument boundary detectionzero-shot LLM evaluationmulti-instance promptinginference cost efficiencyvision-language modelssummarization evaluationLLM-as-a-judgemetric saturationsemantic scaffoldfact preservationQA-based evaluation

Highlights

MSBD: batching page-boundary decisions into one LLM call
On the 2026 arXiv page stream segmentation work (last-author position), MSBD groups multiple split boundary decisions into a single LLM call, holding accuracy at a model-specific knee while lowering inference cost; Evolve judges the approach plausible but notes the evidence is single-run without significance checks.
Semantic scaffold metrics that survive LLM-judge saturation
The 2026 summarization evaluation paper shows standard LLM-as-a-judge scores saturating and proposes FPS, QPS, and EPS — fact- and QA-based semantic scaffolds — that keep score spread; Evolve calls them plausible diagnostics while flagging untested validity against human preference.
A cost-and-rigor agenda for LLM-based text processing
Taken together, the two 2026 papers push one theme: make LLM-based document processing and evaluation cheaper and more diagnostic — batching boundary decisions for inference cost efficiency, and designing metrics that resist saturation — while both Evolve verdicts call for stronger statistical validation.

Timeline

2026Releases the arXiv paper 'Splitting Documents at Lower Cost,' introducing MSBD for LLM-based page stream segmentation from the last-author position.
2026Publishes 'Preserving What Matters,' documenting LLM-judge saturation in summarization evaluation and proposing the FPS/QPS/EPS semantic scaffold metrics.
2026OpenAlex snapshot records 2 works in 2026 with 7 distinct coauthors, spanning natural language processing and document AI at 0.5 each.

Papers in the Evolve catalog

Frequent co-authors

Nikhil Reddy Pottanigari×2 Sepideh Kharaghani×2 Ying Zhang×2 Ramin Fahimi×1 Noah Bolger×1 Saverio Vadacchino×1 Alejandro Posada×1

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.

Bibliometrics come from an OpenAlex snapshot only; h-index and i10-index are recorded as 0 and the output and impact dimensions are unscored, consistent with an unresolved author identity.
No institution, country, or ORCID is available; 'Ying Zhang' is a common name and this profile is name-resolved from public bibliographic data.
The evidence base is two 2026 arXiv papers reviewed by Evolve; the verdicts note single-run evidence and untested validity against human preference.
Collaboration and activity scores derive from a very small sample (2 recent works, 7 distinct coauthors) and will shift as the record grows.