University of Melbourne NLP researcher auditing LLM resume screening for presentation sensitivity and demographic bias
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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—
No verified output data is available beyond the two reviewed 2026 arXiv papers; this axis measures total scholarly production — publication counts, venues, and volume over time — and cannot be scored from the current snapshot.
Impact—
Impact cannot be assessed: no citation counts, h-index, or venue-level influence data are provided. This axis would capture how widely the two LLM screening audits are cited and taken up by the hiring-audit community.
Focus0/100
The focus score is 0. The field distribution splits evenly at 0.5 and 0.5 across two labels — 'natural language processing (LLM evaluation / algorithmic hiring)' and its Chinese-labeled fairness-audit counterpart — and the radar treats this as unfocused. In substance the two 2026 papers address one question, LLM-based resume screening, from evaluation and bias-audit angles, so the low score likely reflects labeling rather than genuinely scattered research.
Collaboration14/100
The collaboration score is 14, low on the 0–100 radar. The bibliometric snapshot records exactly 1 distinct coauthor, Qiangju Chen, who appears as the frequent collaborator across the reviewed set. Both 2026 papers place Yang Xiao in the last author position without corresponding-author status, suggesting a small two-person working pattern in a supporting senior role rather than a large lab network; the sample of two papers limits what this score can show.
Activity—
Activity data is insufficient: the snapshot offers no publication-frequency or per-year volume series. This axis would measure publishing cadence — papers per year and sustained output over time — beyond the two 2026 entries, and is left unscored.
Reproducibility38/100
The reproducibility score is 38, computed as the mean readiness over 2 papers. The 2026 benchmark-audit paper is rated medium — its controlled matched-pair design helps replication, though small scale limits reach — while the demographic-leakage paper is rated low, with Evolve flagging gaps in stimulus construction and definitions. The average of 38 signals that methods sections exist but would need released artifacts and tighter protocols before independent reruns.
Research summary
Yang Xiao is an early-career researcher profiled at The University of Melbourne whose reviewed work converges on one applied question: whether large language model–based resume screening behaves consistently and fairly. Two 2026 arXiv papers, both with Yang Xiao as last author and neither as corresponding author, make up the entire Evolve catalog for this profile. The first is a controlled benchmark audit reporting that LLM resume screeners flip roughly 29–46% of matched decisions when résumés are rewritten in competence-preserving ways, implicating presentation and prompt-format sensitivity; Evolve rates its reproducibility readiness medium. The second probes demographic leakage, reporting a claimed 0.757 ethnicity-recovery score from de-identified, language-preserved text, and argues that verdict format drives audit outcomes; its reproducibility readiness is rated low, with stimulus-construction and definitional gaps flagged. The two papers split 0.5/0.5 across two field labels — LLM evaluation/algorithmic hiring and LLM fairness/bias auditing — and the radar focus score is 0. The collaboration score is 14, with one distinct coauthor, Qiangju Chen. Output, impact, and activity lack sufficient data and are unscored. Because 'Yang Xiao' is a common name and no ORCID is linked, this profile is name-resolved from public bibliographic data.
Collaboration pattern
Collaboration in the reviewed record is minimal: exactly 1 distinct coauthor, Qiangju Chen, who is also the sole frequent co-author, and a radar collaboration score of 14. Both papers put Yang Xiao in the last author position without corresponding status — a consistent supporting-author slot within what looks like a compact two-person team. The two-paper sample is small, so the pattern should be read as an early-stage snapshot rather than a stable collaboration profile.
Research keywords
Highlights
LLM resume screeners flip ~29–46% of matched decisions under competence-preserving rewrites
De-identified résumés still leak demographics, and verdict format shapes the audit
Timeline
- 2026Posts a controlled benchmark audit on arXiv showing LLM resume screeners flip roughly 29–46% of matched decisions under competence-preserving rewrites, with medium reproducibility readiness.
- 2026Publishes the demographic-leakage study on arXiv, reporting a claimed 0.757 ethnicity-recovery score from de-identified, language-preserved résumé text and arguing verdict format drives audit outcomes, with low reproducibility readiness.
- 2026The two arXiv papers together establish a focused agenda auditing LLM-based hiring screens from evaluation and fairness angles, with Yang Xiao as last author on both.
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 only; no ORCID is linked, and 'Yang Xiao' is a common name, so the profile is name-resolved and may merge or split identities.
- Only the two Evolve-reviewed 2026 arXiv papers inform the narrative; output, impact, and activity axes lack underlying data and remain unscored.
- Career facts beyond the stated affiliation 'The University of Melbourne' — country, positions, and history — are absent from the input data.
- The focus score of 0 rests on a 0.5/0.5 split across two field labels from only two papers, so it is highly sensitive to additional records.