Audio-language models for telecom fraud detection: benchmark building and frozen-weight skill optimization
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—
Not scored: the snapshot lacks sufficient publication-count and venue data to measure output. This axis would track annual paper volume, venue tiers, and preprint-to-publication conversion for this author.
Impact—
Not scored: no citation counts, h-index, or downstream-usage data are available in the snapshot. This axis would measure citations to the two 2026 papers and adoption of the TeleAntiFraud benchmarks by other groups.
Focus0/100
The recorded focus score is 0, and the underlying field distribution explains the tension: the fieldDist splits evenly (0.5/0.5) between two labels that both describe audio-language models for fraud detection. In substance, both reviewed 2026 papers sit in one narrow niche — telecom audio anti-fraud benchmarking and frozen-weight adaptation — so topical concentration looks high even though the radar's focus score is 0.
Collaboration58/100
The collaboration score is 58, based on 17 distinct coauthors in the snapshot. Twelve recur in the frequent-coauthor list — including Huiyuan Liu, Shun Zhang, and Qifan Wang — suggesting a stable core team. On both 2026 papers Peng Chen is the last author but not the corresponding author, consistent with a senior, group-anchored position within a mid-sized collaboration network.
Activity—
Not scored: the snapshot provides no year-by-year publication record, so research tempo cannot be computed. This axis would track publishing cadence, recency of outputs, and sustained engagement across the 2026 arXiv record.
Reproducibility78/100
Reproducibility is the strongest scored axis: a mean of 78 across the 2 reviewed papers. Evolve rates TeleAntiFraud 2.0's reproducibility readiness as high — a refreshable benchmark with published design and collapse-aware analysis — and FRAUDSkill as medium. The pattern suggests benchmark artifacts are more reusable than the optimization pipeline, whose gains depend heavily on structured post-processing.
Research summary
Peng Chen is a speech and audio machine learning researcher whose reviewed work centers on audio-language models applied to telecom fraud detection. The Evolve catalog holds two 2026 arXiv papers, on both of which Peng Chen is the last author (non-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. The benchmark paper is rated high for reproducibility readiness, with near-domain negatives and collapse-aware analysis of shortcut learning, though its data is synthetic-only and human validation is thin; the method paper reports a +31.96 Macro-F1 gain on the TeleAntiFraud benchmark, which Evolve's ablation reading attributes mostly to structured post-processing rather than skill search. Bibliometrically, the snapshot records 17 distinct coauthors, 12 of whom recur as frequent collaborators including Huiyuan Liu, Shun Zhang, and Qifan Wang, yielding a collaboration score of 58. The reproducibility score of 78 (mean over the 2 reviewed papers) is the strongest available signal, while output, impact, and activity axes lack sufficient data and are unscored. The recorded focus score is 0 even though both papers sit in the same audio anti-fraud niche. No institution, country, or ORCID is attached, so the profile is name-resolved from public bibliographic data.
Collaboration pattern
The snapshot records 17 distinct coauthors, with a recurring core of 12 frequent collaborators — including Huiyuan Liu, Shun Zhang, Qifan Wang, Yifan Wang, and Yijin Zhou — giving a collaboration score of 58. On both 2026 papers Peng Chen is the last author but not the corresponding author, a pattern consistent with senior, group-anchored positioning in a stable, team-based collaboration rather than solo or widely dispersed work.
Research keywords
Highlights
TeleAntiFraud 2.0: refreshable audio benchmark for telecom fraud detection
FRAUDSkill: frozen-weight skill optimization, +31.96 Macro-F1 with a caveat
Timeline
- 2026Released TeleAntiFraud 2.0 on arXiv: a refreshable, profile-grounded, audio-based benchmark for telecom fraud detection, with near-domain negatives and collapse-aware analysis.
- 2026Published FRAUDSkill on arXiv: structured frozen-weight skill optimization reporting a +31.96 Macro-F1 gain for audio anti-fraud detection.
- 2026Appeared as last (non-corresponding) author on both reviewed papers, within a network of 17 distinct coauthors.
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 dimension scores come from a public bibliometric snapshot; output, impact, and activity axes lack sufficient data and are left unscored.
- "Peng Chen" is a common name; no institution, country, or ORCID is attached, so this profile is name-resolved from public bibliographic data and may not represent a single individual.
- Only 2 papers are in the reviewed catalog, both 2026 arXiv preprints with last-author (non-corresponding) position; venue and peer-review status are unknown.
- Content judgments (gains attributed to post-processing, synthetic-only data, thin human validation) reflect Evolve's editorial verdicts on these two papers, not independent verification.