Xu Yang

Minjiang University

Unclaimed · AI-structured profileOpenAlex

Xu Yang: NLP & LLM-security researcher at Minjiang University — agent privacy leakage and criminal-profiling benchmarks

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—

Output is not yet scored: no publication counts, venue list, or authorship-position breakdown is available for this profile. This axis would measure the volume, venue quality, and consistency of Xu Yang's research output over time.

Impact—

Impact is not yet scored: no citation counts, h-index, or per-paper citation trajectory is available. This axis would measure how widely Xu Yang's work is cited and taken up by other researchers.

Focus0/100

The focus score is 0, derived from a field distribution split exactly 50/50 between natural language processing (legal and criminological NLP) and AI security & privacy (LLM agents). The two catalog papers sit one in each area, so no single dominant field emerges yet. Both, however, share an LLM-evaluation core — benchmarking and stress-testing black-box models in sensitive, high-stakes settings — suggesting a thematic bridge rather than random dispersion.

Collaboration61/100

Collaboration scores 61, based on 20 distinct coauthors recorded in the bibliometric snapshot. The recurring roster includes Yutong Yao, Yanjie Cao, Guanhua Chen, Junchao Wu, Zeyu Wu, Lidia S. Chao, Derek F. Wong, Tao Huang, Guosen Wu, Guolong Zheng, and Jiayang Meng. Both catalog papers are multi-author efforts with Xu Yang in a middle authorship position and not the corresponding author, indicating active team-based research rather than solo work.

Activity—

Recent activity is not yet scored: the data snapshot lacks per-year publication counts and submission cadence. This axis would measure how actively Xu Yang is publishing right now.

Reproducibility55/100

Reproducibility scores 55, an average across 2 papers with Evolve reproducibility reviews, both rated medium readiness. Evolve's verdicts note concrete gaps: CIPL's results are coherent across 5 providers but lack statistical testing, and the criminal-profiling benchmark has no significance tests and no main-set reliability check. The pattern suggests careful framework and benchmark design whose empirical claims would be strengthened by formal statistical validation.

Research summary

Xu Yang is a researcher at Minjiang University whose visible work sits at the junction of natural language processing and AI security. The Evolve catalog holds two 2026 arXiv papers, one in each of the fields the bibliometric snapshot splits 50/50: legal and criminological NLP, and AI security & privacy for LLM agents. The first, CIPL, formalizes channel-aware, recoverable privacy leakage in black-box LLM agents, with results Evolve finds coherent across five providers; the second, Before the Arrest, is a five-country benchmark testing LLMs on criminal profiling from incomplete evidence, praised for strong design and honest limits. In both, Xu Yang appears in a middle authorship position and is not the corresponding author, consistent with a team-based working mode backed by 20 distinct coauthors and a collaboration dimension score of 61. Reproducibility scores 55, the mean across 2 reviewed papers, both rated medium readiness, with Evolve noting in each case an absence of statistical significance testing and, for the benchmark, no main-set reliability check. Output, impact, and activity dimensions are unscored because their underlying data is not yet available; this should therefore be read as an early-stage profile built on a small, recent evidence base rather than a mature publication record.

Collaboration pattern

The collaboration dimension scores 61, drawing on 20 distinct coauthors. Recurring names include Yutong Yao, Yanjie Cao, Guanhua Chen, Junchao Wu, Zeyu Wu, Lidia S. Chao, Derek F. Wong, Tao Huang, Guosen Wu, Guolong Zheng, and Jiayang Meng. On both catalog papers Xu Yang is a middle-position, non-corresponding author, so the visible pattern is contribution within multi-author teams rather than independent leadership; no seniority or role structure can be inferred from the available data.

Research keywords

LLM agentsprivacy leakageblack-box evaluationRAG privacytool-use securitymemory extractionLLM benchmarkcriminal profilingabductive reasoninglegal NLPhomicide case analysissentence prediction

Highlights

CIPL: formalizing recoverable privacy leakage in LLM agents
The 2026 arXiv paper introduces a channel-aware framework that separates exposure from recovery for black-box agent privacy leakage, covering RAG privacy, tool-use security, and memory extraction. Evolve finds the results coherent across 5 providers but flags the absence of statistical testing; reproducibility readiness is rated medium.
Before the Arrest: a five-country LLM criminal-profiling benchmark
The second 2026 paper benchmarks LLMs on criminal profiling from incomplete evidence across five countries, covering abductive reasoning, homicide case analysis, and sentence prediction within legal NLP. Evolve praises the novel design and honest limits while noting no significance tests and no main-set reliability check.
A shared black-box evaluation playbook across two sensitive domains
Both reviewed works apply the same evaluation instinct — structured stress-testing of black-box LLMs in high-stakes settings — linking criminological NLP with agent security. In each, Xu Yang contributes from a middle authorship position, and Evolve rates reproducibility as medium, with a mean score of 55 across the 2 papers.

Timeline

2026CIPL, a channel-aware framework for recoverable privacy leakage in LLM agents, appears on arXiv and enters the Evolve catalog with medium reproducibility readiness.
2026Before the Arrest, a five-country benchmark of LLM criminal profiling from incomplete evidence, appears on arXiv and enters the Evolve catalog with medium reproducibility readiness.
2026The bibliometric snapshot records 20 distinct coauthors, a 50/50 field split between legal/criminological NLP and LLM-agent security, a collaboration score of 61, and a reproducibility mean of 55 across the 2 reviewed papers.

Papers in the Evolve catalog

Frequent co-authors

Yutong Yao×1 Yanjie Cao×1 Guanhua Chen×1 Junchao Wu×1 Zeyu Wu×1 Lidia S. Chao×1 Derek F. Wong×1 Xu Yang×1 Tao Huang×1 Guosen Wu×1 Guolong Zheng×1 Jiayang Meng×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.

Dimension scores are computed from public bibliometric data only; no citation counts, h-index, or funding information was provided for this profile.
Output, impact, and activity axes are unscored because their underlying data is not yet available; conclusions rest on a small, recent evidence base (2 arXiv papers from 2026).
Both catalog entries are arXiv preprints; peer-review status and final publication venues are not reflected in the data.
'Xu Yang' is a common name: this profile is name-resolved from public bibliographic data, and the frequent-coauthor list contains an entry matching the profile's own name, so some records may mix researchers sharing the name.