Document AI researcher advancing cost-aware, rigorous LLM evaluation for document QA, segmentation, and summarization
3
papers here
3
reviewed by Evolve
0
first-author
0
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: this profile draws on three reviewed arXiv papers only, without a full publication count or venue record. This axis would measure total scholarly output over time, including preprints, peer-reviewed articles, and other indexed works.
Impact—
Citation-based impact cannot be scored: no citation counts, h-index, or uptake data are present in the input. This axis would measure how widely the work is cited and adopted by other researchers within and beyond document AI.
Focus0/100
Focus scores 0 because cataloged output is spread evenly across three subfields, each at one third: natural language processing, page stream segmentation, and multimodal document AI. Substantively the spread is narrow — all three sit inside document-oriented NLP — but the even distribution yields no dominant field. As the catalog grows, a main line may concentrate this score.
Collaboration41/100
Collaboration scores 39 from a modest base: six distinct coauthors — Nikhil Reddy Pottanigari, Sepideh Kharaghani, Saverio Vadacchino, Alejandro Posada, Ramin Fahimi, and Noah Bolger — appear across the three cataloged papers. Zhang takes the last author position on all three without corresponding-author role, consistent with a senior seat in a small, stable team rather than an extended network.
Activity—
Activity tempo cannot be scored: the record is limited to three 2026 arXiv entries, with no multi-year publication history or submission-cadence metadata. This axis would measure research pace over time, such as output per year and gaps between releases.
Reproducibility55/100
Reproducibility scores 55, averaged over three reviewed papers that each carry a medium readiness rating. The work is controlled and honest — batching boundary decisions into one LLM call, comparing text versus pixel inputs, proposing scaffold metrics — but Evolve's verdicts repeatedly note single-run evidence without significance testing, which caps this score.
Research summary
Ying Zhang is a researcher working at the intersection of natural language processing and document AI, whose cataloged output concentrates on how large language and vision-language models read, segment, and evaluate documents. Three 2026 arXiv papers, all with Zhang in the last author position, define the record. The first compares text, OCR-derived text, and page-image inputs for multimodal document question answering, reporting that page images beat text on accuracy within a 50-page cap while incurring growing latency cost. The second introduces MSBD, a method that batches page-boundary decisions into a single LLM call for page stream segmentation, holding accuracy to a model-specific knee while lowering inference cost. The third diagnoses metric saturation in LLM-as-a-judge summarization evaluation and proposes semantic scaffold metrics — FPS, QPS, and EPS — designed to preserve score spread. A common thread is rigorous, cost-aware evaluation methodology: measuring what models actually retain at what computational price. Evolve's verdicts describe the work as controlled and honest, while consistently flagging single-run evidence without significance testing; reproducibility readiness is rated medium across all three papers (mean 55). The collaboration footprint is compact: six distinct coauthors. Because output, impact, and activity metrics are missing, this profile should be read as an early-stage record built from the reviewed catalog alone.
Collaboration pattern
All three cataloged papers list Zhang as last author (none corresponding), working with a compact, recurring group of six coauthors: Nikhil Reddy Pottanigari, Sepideh Kharaghani, Saverio Vadacchino, Alejandro Posada, Ramin Fahimi, and Noah Bolger. The pattern suggests a small, stable team producing evaluation- and efficiency-focused document AI work on arXiv in 2026. No cross-institution or industry partnership can be inferred from the available data.
Research keywords
Highlights
Text vs. pixels vs. both: input representations for multimodal document QA
MSBD: batching page-boundary decisions into one LLM call for cheaper segmentation
Semantic scaffolds beyond LLM-judge saturation in summarization evaluation
Timeline
- 2026Released 'Text, Pixels, or Both?' on arXiv, evaluating input representations for multimodal document QA as last author.
- 2026Published MSBD for LLM-based page stream segmentation, batching boundary decisions into a single LLM call.
- 2026Proposed semantic scaffold metrics (FPS/QPS/EPS) addressing LLM-judge saturation in summarization evaluation.
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.
- 'Ying Zhang' is a common name: this profile is name-resolved from public bibliographic data and the three reviewed arXiv papers; ORCID and affiliation are absent, so identity is not independently verified.
- Bibliometric scores come from OpenAlex-derived data and the Evolve-reviewed catalog only; output, impact, and activity axes lack sufficient data and are left unscored.
- The record covers three 2026 arXiv preprints; earlier, peer-reviewed, or non-arXiv work may be missing.
- Collaboration (39), focus (0), and reproducibility (55) scores reflect only the three reviewed papers, not a full career record.