Audio-language models for telecom fraud detection: benchmark builder and frozen-weight adaptation researcher
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
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—
Output metrics — publication volume, venue types, and output rate over time — are not yet available in the underlying data. This axis would measure total scholarly output; only two reviewed 2026 preprints are currently in scope, so the axis is unscored.
Impact—
Impact metrics such as citations, h-index, and downstream reuse are not available yet. This axis would measure how far the benchmark and method papers propagate through the fraud-detection and speech communities; it remains unscored.
Focus0/100
The focus axis is scored 0. Across only two cataloged papers, the raw field distribution splits evenly (0.5/0.5) between 'speech processing and NLP (audio-language model benchmarking for fraud detection)' and 'speech and audio machine learning (audio-language models for fraud detection)'. The two labels overlap substantively, so a zero score most plausibly reflects an insufficient sample for measuring concentration; no stable topical-focus signal can be established from this evidence base.
Collaboration58/100
Collaboration scores 58, computed from 17 distinct coauthors across the reviewed work. Twelve collaborators recur, including Huiyuan Liu, Shun Zhang, Qifan Wang, and Yifan Wang, indicating a stable team working on one agenda. On both 2026 papers Zhou appears in a middle, non-corresponding position, consistent with contributing inside a multi-author group rather than leading an independent line.
Activity—
Activity metrics — recent publication cadence and yearly output distribution — cannot be computed from a snapshot containing only 2026 papers. This axis would measure ongoing research pace; there is no multi-year signal yet, so it stays unscored.
Reproducibility78/100
Reproducibility scores 78, the mean over the two reviewed papers. TeleAntiFraud 2.0 is rated high reproducibility readiness — a refreshable, profile-grounded, audio-based benchmark with collapse-aware analysis — while FRAUDSkill is rated medium; Evolve notes its ablations attribute most of the +31.96 Macro-F1 gain to structured post-processing rather than the skill search itself.
Research summary
Yijin Zhou is an early-stage researcher whose reviewed work concentrates on audio-language models for telecom fraud detection. The Evolve catalog contains two 2026 arXiv papers, in both of which Zhou holds a middle author position (not corresponding): TeleAntiFraud 2.0, a refreshable, profile-grounded, audio-based benchmark for telecom fraud detection, and FRAUDSkill, a structured frozen-weight skill optimization method for audio anti-fraud detection. Together they outline a coherent agenda: build benchmark infrastructure for audio fraud detection, then optimize model behavior on top of it. Evolve's review notes TeleAntiFraud 2.0's strong design and collapse-aware analysis — including near-domain negatives and shortcut-learning controls — while flagging synthetic-only data and thin human validation; FRAUDSkill's reported +31.96 Macro-F1 gain is attributed by its own ablations mostly to structured post-processing rather than skill search. Bibliometrics show 17 distinct coauthors and an even split between two closely related field labels, with a mean reproducibility score of 78 across the two reviewed papers. Output, impact, and activity metrics are not yet available, so this is an early-stage profile with a narrow but coherent evidence base.
Collaboration pattern
Zhou's reviewed work is team-based: 17 distinct coauthors appear across the catalog, 12 of whom recur frequently — including Huiyuan Liu, Shun Zhang, Qifan Wang, Yifan Wang, Yuxi Zhao, Chengxian Hu, and Peng Chen. The collaboration score is a moderate 58, and on both 2026 papers Zhou sits in a middle, non-corresponding position, consistent with contributing within a larger group on a shared audio anti-fraud agenda.
Research keywords
Highlights
TeleAntiFraud 2.0: refreshable audio benchmark for telecom fraud detection
FRAUDSkill: frozen-weight skill optimization for audio anti-fraud detection
A paired benchmark-plus-method agenda
The two 2026 papers combine evaluation infrastructure (TeleAntiFraud 2.0) with an optimization method (FRAUDSkill) on the same fraud-detection task, all produced within a 17-person collaboration network.
Timeline
- 2026Co-authored TeleAntiFraud 2.0 (arXiv), a refreshable, profile-grounded, audio-based benchmark for telecom fraud detection, rated high reproducibility readiness by Evolve.
- 2026Co-authored FRAUDSkill (arXiv), applying structured frozen-weight skill optimization to audio anti-fraud detection, with a reported +31.96 Macro-F1 gain largely attributed by ablations to structured post-processing.
- 2026Accumulated 17 distinct coauthors across the two reviewed papers, with 12 collaborators recurring frequently around the audio anti-fraud agenda.
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 as provided in this snapshot; output, impact, and activity axes lack sufficient data and are unscored.
- No institution, country, or ORCID metadata was available; the profile is name-resolved from public bibliographic data, and 'Yijin Zhou' is a common name — verify identity before reuse.
- The evidence base is two 2026 arXiv preprints reviewed by Evolve; venue acceptance and citation data are outside this snapshot.
- All figures (e.g., +31.96 Macro-F1, 17 coauthors, reproducibility 78) are quoted verbatim from the provided data and Evolve review notes, not independently verified.