NLP benchmark builder: cross-lingual Chinese buzzword understanding and web-agent safety evaluation
2
papers here
2
reviewed by Evolve
1
first-author
0
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—
Not yet measurable: Evolve lacks verified publication-count and venue data for this researcher. This axis would track total output volume, venue tiers, and first/corresponding-author share; with only two reviewed arXiv papers, no score is assigned.
Impact—
Insufficient data: no citation counts, h-index, or downstream-adoption evidence are available for this profile. This axis would measure citation impact and real-world uptake of the two benchmarks; no score is estimated.
Focus0/100
The focus axis scores 0, while the raw field distribution splits evenly (0.5/0.5) between general natural language processing and NLP for content moderation and web-agent safety evaluation. Substantively, both reviewed papers are closely related 2026 LLM benchmarks on Chinese-language obfuscation, cultural pragmatics, and platform-risk investigation, so the zero score appears to track the coarse two-way field split rather than genuinely scattered interests. Read this axis together with the underlying field distribution.
Collaboration62/100
The collaboration axis scores 62. The underlying record shows 22 distinct co-authors and a list of 12 frequent collaborators, including Shun Zhang, ZhuoXin Liu, and Zhiming Ma. Both Evolve-reviewed papers are multi-author efforts, and the first-author role on CIBuzzBench shows capacity to lead team-based benchmark construction. The mid-range score reflects a moderately broad, recurring author network around two 2026 benchmark papers; longer-horizon patterns cannot be confirmed from this snapshot alone.
Activity—
Insufficient data: publication cadence, yearly volume, and recent-activity trends cannot be computed from a snapshot containing only 2026 arXiv records. This axis would measure sustained research pace over multiple years; no score is assigned.
Reproducibility78/100
The reproducibility axis scores 78, a mean computed across two papers (reproMean 78, reproPapers 2). CIBuzzBench is rated high for reproducibility readiness, supported by strong human-verified annotation of its 3,001 entries, while RiskChainBench is rated medium, with Evolve noting that synthetic data and single-run evaluation limit generalizability. A high-seventies mean on a two-paper base is a solid early signal, though it should be treated as provisional until more artifacts are reviewed.
Research summary
Yifan Wang, affiliated with China People's Public Security University, is an early-documented natural language processing researcher whose Evolve-reviewed work centers on evaluation benchmarks for large language models in Chinese-language and safety-critical settings. Two 2026 arXiv papers define the visible record. The first-authored CIBuzzBench is described by Evolve as the first benchmark (3,001 entries, 3 tasks) for Chinese-to-English internet-buzzword understanding, with strong human-verified annotation and high reproducibility readiness, though contamination risk is unexamined. The second, RiskChainBench, is a two-stage benchmark connecting obfuscated platform-message restoration with evidence-grounded web-agent risk investigation; Evolve rates its design highly but flags synthetic data and single-run evaluation as limits on generalizability, with medium reproducibility readiness. The radar shows a collaboration score of 62, built on 22 distinct co-authors and 12 recurring collaborators, and a reproducibility mean of 78 across two papers. The field distribution splits evenly between general NLP and NLP for content moderation and web-agent safety evaluation, indicating a benchmark-focused specialization at the intersection of language, culture, and platform safety. Output, impact, and activity axes remain unmeasured, so this should be read as an early-stage, benchmark-centric profile.
Collaboration pattern
Yifan Wang works in multi-author teams: the reviewed record lists 22 distinct co-authors, with 12 recurring collaborators, including Shun Zhang, ZhuoXin Liu, Zhiming Ma, Ying Zhang, and Mengzheng Yang. Both 2026 benchmark papers are collective builds, with Wang first-authoring CIBuzzBench and contributing mid-authorship to RiskChainBench. The pattern suggests a stable core group producing evaluation benchmarks together, within a collaboration score of 62; broader cross-institution ties are not visible in the current data.
Research keywords
Highlights
CIBuzzBench: first benchmark for Chinese-to-English buzzword understanding
RiskChainBench links obfuscated message restoration to web-agent risk investigation
Timeline
- 2026First-authored CIBuzzBench (arXiv, 2026), a 3,001-entry, three-task benchmark for cross-lingual understanding of Chinese internet buzzwords.
- 2026Co-authored RiskChainBench (arXiv, 2026), a two-stage benchmark for obfuscated platform-message restoration and evidence-grounded web investigation.
- 2026Both 2026 releases are evaluation benchmarks, receiving Evolve reproducibility-readiness ratings of high and medium across two papers.
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 scores are derived from OpenAlex-based public data; output, impact, and activity axes lack sufficient data and are unscored.
- Name resolved from public bibliographic data: 'Yifan Wang' is a common Chinese name and no ORCID is attached, so identity matching carries some uncertainty.
- Scope covers only two Evolve-reviewed arXiv papers from 2026; no citation, venue, award, or career-history information is included.
- The reproducibility mean (78) is computed from only two papers, so it should be read as provisional.