First author of BenchCompass, a 306-item payment-domain LLM benchmark; works on LLM evaluation and financial NLP.
1
papers here
1
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-21) 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: the current data scope covers a single reviewed 2026 paper, so publication volume, venue trajectory, and productivity trends cannot be measured. This axis would aggregate paper counts and output regularity over time.
Impact—
Impact is not yet scored because citation counts and downstream usage are unavailable in the provided data. This axis would measure citations, adoption of the benchmark, and influence on subsequent payment-domain LLM evaluation work.
Focus100/100
Focus scores at the maximum of 100 because the bibliometric field distribution contains exactly one field — natural language processing / LLM evaluation — covering the reviewed BenchCompass paper. A single-field profile at this stage indicates a tightly concentrated agenda around payment-domain benchmark construction, attack robustness, and LLM-as-a-judge methodology, though with only one cataloged paper the concentration is early evidence rather than an established pattern.
Collaboration50/100
Collaboration scores 50, mid-scale, built on 11 distinct coauthors recorded across the cataloged work. The frequent-coauthor list is broad — including Wei Ren, Xuanwei Hu, Jiawei Luo, Zifan Wang, Xiaoyun Feng, Hui Cai, Lyuxin Xue, Peng Lu, Jianshe Li, Xin Zhang, and Wei Wu — suggesting a team-based working style, while the mid-range score reflects the limited number of cataloged papers over which the network is measured.
Activity—
Activity is not yet scored: with only a 2026 paper in the provided data, there is no basis for measuring publication cadence, recent-year output, or sustained engagement. This axis would track output frequency and recency across the publication record.
Reproducibility100/100
Reproducibility scores 100, a perfect mean computed from exactly one paper: BenchCompass, whose reproducibility readiness Evolve rates as high. The payment-domain benchmark is delivered as a concrete 306-item artifact, which supports the top score; however, with only one assessed paper, this axis reflects a strong single data point rather than a demonstrated track record of consistently reproducible releases.
Research summary
Sijie Dong is an early-stage researcher in natural language processing whose cataloged work centers on LLM evaluation, currently consisting of a single reviewed paper: BenchCompass (arXiv, 2026), on which Dong is first author. The paper contributes a 306-item payment-domain benchmark that separates three evaluation axes — domain knowledge, context use, and robustness to task-input attacks — and argues that benchmark scores should translate into signals for training decisions and harness choices. Evolve's review judges it a solid payment-domain benchmark while noting that small subsets and judge-family overlap temper some claims; its reproducibility readiness is rated high, and it is the single paper behind this profile's reproducibility mean of 100. The bibliometric field distribution contains exactly one field, natural language processing / LLM evaluation, yielding a focus score of 100. Dong's collaboration network spans 11 distinct coauthors — including Wei Ren, Xuanwei Hu, Jiawei Luo, Zifan Wang, and Xiaoyun Feng — producing a mid-range collaboration score of 50. Output, impact, and activity are not yet scored because the underlying data is insufficient, and no institution, country, or ORCID is attached, so this profile is name-resolved from public bibliographic data.
Collaboration pattern
Sijie Dong's recorded collaboration network spans 11 distinct coauthors tied to the BenchCompass work: Wei Ren, Xuanwei Hu, Jiawei Luo, Zifan Wang, Xiaoyun Feng, Hui Cai, Lyuxin Xue, Peng Lu, Jianshe Li, Xin Zhang, and Wei Wu. Eleven collaborators on a single first-authored benchmark paper indicates a team-based effort; the collaboration score of 50 reflects a broad but thinly measured network, since only one paper underlies it.
Research keywords
Highlights
BenchCompass: a 306-item payment-domain LLM benchmark
From scores to signals for training and harness decisions
High reproducibility readiness with acknowledged limits
Timeline
- 2026First-authors BenchCompass on arXiv, a 306-item payment-domain LLM benchmark separating knowledge, context use, and attack robustness.
- 2026Evolve reviews the paper as a solid payment-domain benchmark, noting that small subsets and judge-family overlap temper its claims.
- 2026The paper's reproducibility readiness is rated high, establishing this profile's reproducibility mean of 100 from a single paper.
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.
- All bibliometrics come from the platform's bibliometric snapshot; output, impact, and activity are unscored because the underlying data is insufficient.
- The evidence base is one reviewed paper (BenchCompass, arXiv 2026), so every dimension rests on a single observation.
- No institution, country, or ORCID is attached; this profile is name-resolved from public bibliographic data and may be affected by name ambiguity.
- Reproducibility (mean 100) and focus (100) are each computed from one paper and should be read as early-stage signals, not established track records.