Qiangju Chen

Macquarie University

Unclaimed · AI-structured profileOpenAlex

Macquarie University auditor of LLM résumé screening: presentation sensitivity and demographic leakage in hiring

papers here

reviewed by Evolve

first-author

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—

Unscored: the snapshot holds no publication-count data beyond the two reviewed preprints. With fuller record linkage this axis would measure total output, preprint throughput, and venue spread, showing how much this researcher publishes each year.

Impact—

Unscored: no citation counts, h-index, or downstream-uptake signals appear in the provided data. This axis would measure citations to the two preprints and their uptake in hiring-audit follow-up work, once the bibliographic record matures.

Focus0/100

Radar records focus at 0 on the 0-100 scale. The raw field distribution splits evenly at 0.5/0.5 between two labels — LLM evaluation for algorithmic hiring and LLM fairness/bias auditing — that describe one and the same problem space: auditing large language models used in résumé screening. Both reviewed papers target this single question, so the agenda is highly concentrated and early rather than broad.

Collaboration14/100

Collaboration scores 14, consistent with a very compact working mode: the raw record lists exactly one distinct co-author, Yang Xiao, across both reviewed papers. Chen appears as first author on each, with no corresponding-author role recorded, so the current footprint is a small, tightly scoped two-person collaboration rather than a network.

Activity—

Unscored: the data provides paper years but no continuous publication cadence or timestamped output history. This axis would measure yearly submission rhythm and research continuity; the record currently shows only two arXiv postings in 2026, too thin to trend.

Reproducibility38/100

Reproducibility averages 38 across the two rated papers (reproPapers: 2). The résumé-perturbation benchmark is rated medium — its controlled, O*NET-grounded design supports replication despite small scale — while the de-identification audit is rated low, with gaps in stimulus construction and metric definition flagged before results like the 0.757 ethnicity-recovery figure can be independently reproduced.

Research summary

Qiangju Chen, listed with Macquarie University, works at the intersection of natural language processing evaluation and algorithmic-hiring fairness. The reviewed record comprises two first-authored arXiv preprints, both dated 2026, that audit large language models used to screen résumés. The first, 'Competence-Preserving Resume Perturbations Expose Presentation Sensitivity in LLM Screening', builds a controlled benchmark in which competence-relevant content is held fixed while phrasing changes, and finds that LLM screeners flip roughly 29–46% of matched screening decisions; Evolve judged the O*NET-grounded design sound while noting that small scale limits its reach. The second, 'Beyond the Name', studies demographic leakage in de-identified résumés and evaluation artifacts in LLM bias audits, reporting ethnicity recovery of 0.757 from text descriptions alone under language-constant conditions, and showing that judgment format drives audit conclusions. The radar scores collaboration at 14 (a single distinct co-author, Yang Xiao), focus at 0 with an even 0.5/0.5 field split across two labels of the same LLM-hiring-audit area, and reproducibility at 38 averaged over two papers rated medium and low. Output, impact, and activity remain unscored pending citation and volume data. The result is an early-stage, tightly focused profile on whether LLM hiring tools judge merit or presentation.

Collaboration pattern

The reviewed record shows a tight, two-person pattern: across both first-authored arXiv papers there is exactly one distinct co-author, Yang Xiao, and no corresponding-author role is recorded. Collaboration scores 14 on the radar, consistent with an early-stage researcher working in a small core team; no institutional or cross-group collaboration networks are visible in the data yet.

Research keywords

LLM-based resume screeningpresentation invariancecompetence-preserving perturbationsbenchmark auditO*NET occupational dataprompt-format sensitivityde-identificationdemographic inferencecounterfactual evaluationLLM-as-a-judgealgorithmic hiring fairnessbias auditing

Highlights

LLM screeners flip 29–46% of matched decisions on competence-preserving rewrites
In this first-authored arXiv benchmark, rewriting résumés while keeping competence content fixed causes LLM screeners to overturn roughly 29–46% of matched screening decisions, exposing presentation sensitivity; Evolve judged the O*NET-grounded design sound but noted that small scale limits the finding's reach.
De-identified résumés still leak ethnicity (0.757 recovery)
'Beyond the Name' reports that, with language held constant, demographic attributes remain recoverable from de-identified résumé text — ethnicity recovery at 0.757 — showing that removing names alone does not strip demographic signal from LLM screening inputs.
Audit conclusions hinge on evaluation artifacts
The same study finds that judgment format drives audit outcomes: how the LLM judge is prompted and what it scores changes conclusions. It flags gaps in stimulus construction and definition — a caution for anyone running LLM-based bias audits.

Timeline

2026Posts first-authored arXiv preprint 'Competence-Preserving Resume Perturbations Expose Presentation Sensitivity in LLM Screening', a controlled benchmark grounding perturbations in O*NET occupational requirements.
2026Publishes 'Beyond the Name: Demographic Leakage in De-Identified Résumés and Evaluation Artifacts in LLM Bias Audits', reporting 0.757 ethnicity recovery and format-driven audit outcomes.
2026Both reviewed first-author papers appear in the same year, establishing a concentrated agenda on LLM hiring-fairness auditing with co-author Yang Xiao.

Papers in the Evolve catalog

2026-09-15first author
Competence-Preserving Resume Perturbations Expose Presentation Sensitivity in LLM Screening
2026-09-15first author
Beyond the Name: Demographic Leakage in De-Identified Résumés and Evaluation Artifacts in LLM Bias Audits

Frequent co-authors

Yang Xiao×2

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 scores derive from a very small reviewed record — two arXiv preprints, both 2026 — so dimensions may change substantially as the catalog grows.
Output, impact, and activity axes are unscored because citation counts, h-index, and volume history are absent from the input snapshot.
No ORCID or country was provided; this profile is name-resolved from public bibliographic data, and 'Qiangju Chen' may match more than one researcher.
Coverage is limited to papers reviewed by Evolve; other publications, if any, are not reflected in these dimensions.