Fusheng Jin

Beijing Institute of Technology

Unclaimed · AI-structured profileORCID OpenAlex

Fusheng Jin — NLP researcher at Beijing Institute of Technology focused on low-resource and cross-domain NER with LLMs

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-13) 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—

No output score yet: the snapshot lacks publication counts, venue tiers, and total paper volume. This axis would measure cumulative research productivity over time; available data covers only the two reviewed 2026 arXiv preprints, too thin to score.

Impact—

No impact score yet: citation counts, citation trajectories, and venue influence are missing from the data. This axis would measure how widely the NER papers and earlier work have been adopted and cited — not scoreable from the current snapshot.

Focus100/100

Focus scores 100 out of 100, and the raw field distribution holds exactly one field: natural language processing. Both cataloged papers sit on a single sub-line — low-resource and cross-domain named entity recognition with large language models (URA-NER, 2026; the entity-type-confusion study, 2026). It is a deliberately narrow agenda of retrieval augmentation and reasoning enhancement for NER rather than scattered topics; this concentrates expertise, though it leaves the agenda sensitive to shifts in the LLM-for-NER wave.

Collaboration32/100

Collaboration scores 32 out of 100, based on a raw count of 4 distinct co-authors in the snapshot. The pattern is a small, stable core team: Jingyu Wang and Shijie Wu recur across both 2026 preprints, with Jin in the senior last-author slot (corresponding status not flagged). This compact setup supports fast iteration on shared NER and LLM infrastructure, but the narrow author pool offers no evidence yet of external or cross-institution collaboration.

Activity—

No activity score yet: yearly publication cadence and recent-output trends are not in the provided data, so current momentum cannot be scored. This axis would measure sustained publishing activity across years rather than the single-2026 snapshot now available.

Reproducibility38/100

Reproducibility scores 38, computed as a mean across 2 scored papers (reproMean 38, reproPapers 2). Readiness splits by paper: medium for URA-NER, whose ablations Evolve judges good but which lacks significance tests or leakage controls, and low for the cross-domain type-confusion paper, whose all-domain SOTA claim conflicts with its own Table 2. Reporting statistical rigor and releasing code and data would most directly lift this axis.

Research summary

Fusheng Jin is a researcher at Beijing Institute of Technology working in natural language processing, with a recent, tightly focused line of research on named entity recognition (NER) under low-resource and cross-domain conditions using large language models. The Evolve catalog lists two 2026 arXiv preprints on which Jin appears as last author: URA-NER, a unified retrieval-augmented framework combining retrieval alignment and uncertainty reduction for low-resource NER, which lifts Qwen2.5-14B to state-of-the-art CrossNER results in a training-free manner; and a cross-domain NER study that quantifies entity type confusion (MCQM) and adds bidirectional LLM reasoning (PBRC), reporting strong CrossNER gains. Evolve's review credits solid ablations in the first paper while noting the absence of significance tests or leakage controls, and flags that the second paper's SOTA-on-all-domains claim overreaches relative to its own Table 2. Bibliometric indicators show extreme focus — a single field, natural language processing, accounts for the record — a compact collaboration network of 4 distinct co-authors anchored by recurring partners Jingyu Wang and Shijie Wu, and a reproducibility score of 38 across 2 scored papers. Output, impact, and activity metrics are not yet available in the snapshot.

Collaboration pattern

The snapshot records 4 distinct co-authors, and the same two names — Jingyu Wang and Shijie Wu — recur across both 2026 preprints, suggesting a small, stable core team rather than a broad network. Jin sits in the last (senior) author position on both papers, with corresponding-author status not flagged. This compact pattern supports rapid iteration on shared NER and LLM infrastructure, but leaves external and cross-institution collaboration unproven in the available data.

Research keywords

named entity recognitionlow-resource NLPretrieval-augmented generationin-context learningdemonstration retrievallarge language modelscross-domain NERentity type confusionLLM reasoningknowledge augmentationlow-resource transfer

Highlights

Training-free retrieval lifts Qwen2.5-14B to low-resource NER SOTA
URA-NER is a unified retrieval-augmented framework for low-resource NER that combines retrieval alignment and uncertainty reduction. Evolve's review credits a training-free pipeline that reaches state-of-the-art CrossNER results with Qwen2.5-14B and includes good ablations, while flagging missing significance tests and leakage controls (reproReadiness: medium).
Quantifying entity type confusion in cross-domain NER
A companion 2026 preprint introduces multidimensional quantification of entity type confusion (MCQM) plus a bidirectional LLM reasoning mechanism (PBRC), reporting strong CrossNER gains. Evolve rates reproducibility low, noting that the claim of SOTA on all domains overreaches relative to the paper's own Table 2.
A compact senior-author research agenda
On both cataloged papers Jin appears as last author with the same recurring co-authors, Jingyu Wang and Shijie Wu, offering early evidence of a coherent, senior-led agenda: retrieval-augmented and reasoning-enhanced LLM methods for low-resource and cross-domain NER.

Timeline

2026Publishes URA-NER on arXiv as last author: a unified retrieval-augmented framework for low-resource NER combining retrieval alignment and uncertainty reduction.
2026Publishes a cross-domain NER study introducing multidimensional quantification of entity type confusion (MCQM) and LLM reasoning enhancement (PBRC).
2026Both preprints appear in the same year, consolidating a focused line on low-resource and cross-domain NER with large language models.

Papers in the Evolve catalog

Frequent co-authors

Jingyu Wang×2 Shijie Wu×2 Jingyu Wang×1 Shijie Wu×1

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 and raw metrics come from a single bibliometric snapshot covering only 4 distinct co-authors and 2 reviewed papers; they may not reflect the full publication record.
Output, impact, and activity axes are unscored because citation counts, venue data, and yearly productivity are absent from the input data.
Both cataloged works are 2026 arXiv preprints; no peer-reviewed venue or publication history is present in the data.
Profile is name-resolved from public bibliographic data (ORCID 0000-0002-0609-8833); the name Fusheng Jin may correspond to multiple researchers, and affiliation is taken as given.