Audio-language-model researcher for telecom anti-fraud: refreshable benchmarks and frozen-weight skill optimization.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
5
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-19) 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—
The output dimension is not yet scored: the OpenAlex snapshot for this name-resolved profile is sparse (8 indexed works, 2019-2024, h-index 5) and no complete publication list or venue record is available, so total output volume and its quality cannot be measured reliably yet.
Impact—
Citation-impact data is insufficient for a score. The raw snapshot provides only h-index 5 and i10-index 4 without citation totals or venue-level signals, so this axis would measure citation accumulation and scholarly influence but cannot yet be rated.
Focus0/100
The focus score is 0/100, the lowest radar axis. The OpenAlex topic profile spreads across five clusters — privacy-preserving technologies (6 works), mobile crowdsensing (4), cooperative communication and network coding (2), auction theory (2), and stochastic gradient optimization (2) — while the two 2026 catalog papers concentrate on a newer area, audio-language models for fraud detection. This breadth and recent pivot make the current research focus hard to pin down.
Collaboration58/100
Collaboration scores 58/100. The raw snapshot records 17 distinct coauthors, and 12 recur as frequent collaborators, led by Huiyuan Liu, Shun Zhang, and Qifan Wang. On both 2026 catalog papers Yuxi Zhao is a middle author and not corresponding, suggesting contribution within a larger team rather than lead authorship. The moderate score reflects an active but not yet lead-positioned collaboration network.
Activity20/100
Activity scores 20/100, a low reading. countsByYear shows 1 work in 2019, 1 in 2020, 4 in 2021, nothing indexed in 2022, then 1 each in 2023 and 2024 — eight works over six active years — with recent3yWorks at just 1. The 2026 arXiv papers suggest renewed output, but indexed recent volume remains thin.
Reproducibility78/100
Reproducibility scores 78/100, the strongest radar axis, with a reproMean of 78 computed across 2 reviewed papers. TeleAntiFraud 2.0 is rated high readiness thanks to its refreshable benchmark-oriented design, while FRAUDSkill is medium; Evolve's ablation check on the +31.96 Macro-F1 claim shows the review process interrogates reported gains rather than accepting them at face value.
Research summary
Yuxi Zhao is a researcher at the First Affiliated Hospital of Zhejiang University, China, whose OpenAlex-resolved bibliometric profile spans eight indexed works between 2019 and 2024 (h-index 5, i10-index 4). The Evolve catalog centers on two 2026 arXiv contributions to audio-based telecom fraud detection. TeleAntiFraud 2.0 (middle authorship) proposes a refreshable, profile-grounded, audio-only benchmark built from synthetic dialogue generation with near-domain negatives and collapse-aware analysis; Evolve's review praises the design while flagging synthetic-only data and thin human validation. FRAUDSkill (middle authorship) explores structured frozen-weight skill optimization for audio anti-fraud, reporting a +31.96 Macro-F1 gain on the TeleAntiFraud benchmark that Evolve's ablation reading attributes largely to structured post-processing rather than skill search. Earlier bibliometric topics — privacy-preserving technologies (6 works), mobile crowdsensing (4), cooperative communication and network coding (2), auction theory (2), and stochastic gradient optimization (2) — indicate a broader data-centric background that has since narrowed toward fraud-detection speech AI. The radar shows strong reproducibility (78/100 across two reviewed papers) and moderate collaboration (58/100, 17 distinct coauthors), against low recent activity (20/100, one work in the last three years). Both catalog papers are arXiv preprints with middle, non-corresponding authorship, so leadership claims remain unverified.
Collaboration pattern
Collaboration is team-based: the raw snapshot records 17 distinct coauthors, with 12 frequent collaborators led by Huiyuan Liu, Shun Zhang, and Qifan Wang, alongside Yifan Wang, Yijin Zhou, and others. On both 2026 Evolve-catalog papers Yuxi Zhao sits in the middle of the author list and is not corresponding, indicating contributor rather than lead roles. The collaboration score is 58/100.
Research keywords
Highlights
Refreshable audio benchmark for telecom fraud detection
Frozen-weight skill optimization with a scrutinized +31.96 Macro-F1 gain
Consistent theme: closing shortcut-learning gaps in audio anti-fraud
Both reviewed 2026 papers attack the reliability of audio anti-fraud systems — one through near-domain negatives and prediction-collapse analysis in the benchmark itself, the other through structured output control in frozen-weight adaptation. Across the two reviews the reproducibility mean is 78/100, and neither paper lists Yuxi Zhao as corresponding author.
Timeline
- 2019First indexed work appears in the OpenAlex profile (1 publication that year).
- 2021Most productive indexed year in the profile with 4 works, coinciding with the early topic clusters such as privacy-preserving technologies and mobile crowdsensing.
- 2023Indexed output resumes after the 2022 gap with 1 work.
- 2024Most recent indexed work in the OpenAlex snapshot (1 publication); recent3yWorks stands at 1.
- 2026Two arXiv preprints enter the Evolve catalog — TeleAntiFraud 2.0 and FRAUDSkill — both on audio-language models for telecom fraud detection, with middle authorship.
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.
- Bibliometric metrics (h-index 5, i10-index 4, countsByYear, topics) come from an OpenAlex-resolved snapshot; Zhao is a common Chinese name, so this profile is name-resolved from public bibliographic data only.
- No citation totals, venue list, or complete publication list were provided, which is why the output and impact dimensions are unscored rather than estimated.
- Both catalog papers are 2026 arXiv preprints with middle, non-corresponding authorship; peer-reviewed status and any leadership claims are unverified.
- The early topic clusters (privacy-preserving technologies, mobile crowdsensing) and the recent audio anti-fraud work may reflect different project lines; this profile cannot confirm they form one continuous research program.