Shijie Wu

Beijing Institute of Technology

Unclaimed · AI-structured profileOpenAlex

Beijing Institute of Technology researcher applying retrieval augmentation and LLM reasoning to low-resource NER

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 publication-count or venue-level output data exists in this snapshot, so the output axis is not scored. It would measure cumulative paper count, venue tiers, and the share of first or corresponding-author positions over the researcher's career.

Impact—

Impact cannot be scored: citation counts, h-index, and downstream uptake are absent from the provided data. This axis would track how often the two reviewed NER papers are cited and adopted by subsequent work.

Focus100/100

The focus dimension scores 100/100, and the raw field distribution explains the extreme value: every catalogued paper sits in a single field, natural language processing, and both 2026 papers attack one problem family, named entity recognition, spanning low-resource demonstration retrieval and cross-domain type confusion. That is an unusually narrow, coherent line for an early-stage profile rather than a scattered topic spread.

Collaboration32/100

Scored 32/100, the collaboration dimension reflects a compact network: the raw data lists only 4 distinct coauthors, and Jingyu Wang and Fusheng Jin recur across the Evolve-reviewed work, pointing to a small, stable team at Beijing Institute of Technology. On both 2026 papers Shijie Wu holds a middle author position without corresponding-author role, consistent with a contributing member of that team rather than an independent group lead.

Activity—

Recent-activity data such as yearly publication counts is not present in the snapshot, so the activity axis is unscored. It would measure publication cadence and research continuity across recent years, currently evidenced only by two papers dated 2026.

Reproducibility38/100

The 38/100 score is the mean repro-readiness across 2 assessed papers. URA-NER rates medium: a training-free pipeline with good ablations, though Evolve notes no significance tests or leakage controls. The cross-domain type-confusion paper rates low. Neither paper currently provides the statistical rigor or evaluation controls a high score would require, so independent replication carries moderate risk.

Research summary

Shijie Wu is a researcher affiliated with Beijing Institute of Technology whose Evolve profile is built on two 2026 arXiv preprints in natural language processing, both on named entity recognition (NER) and both listing Shijie Wu in a middle author position without corresponding-author status. The first, URA-NER, is a unified retrieval-augmented framework combining retrieval alignment and uncertainty reduction for low-resource NER; Evolve's review reports that this training-free setup lifts Qwen2.5-14B to state of the art on CrossNER low-resource NER, with good ablations but no significance tests or leakage controls. The second quantifies entity type confusion in cross-domain NER via a multidimensional measure (MCQM) and adds LLM-based bidirectional reasoning (PBRC), reporting strong CrossNER gains, though the review flags a SOTA-on-all-domains claim that overreaches relative to the paper's own Table 2. The radar profile shows an exceptionally narrow focus (100/100, with a single-field distribution in natural language processing), a compact collaboration network (32/100, 4 distinct coauthors, with Jingyu Wang and Fusheng Jin recurring), and a moderate reproducibility score (38/100, the mean across 2 papers rated medium and low). Output, impact, and activity axes lack underlying bibliometric data and are not scored. This is an early-stage, name-resolved profile scoped to Evolve's reviewed catalog.

Collaboration pattern

Shijie Wu's collaboration footprint is small: the raw data lists 4 distinct coauthors, and the frequent-coauthor list is dominated by Jingyu Wang and Fusheng Jin, pointing to a compact, recurring team at Beijing Institute of Technology. On both reviewed 2026 papers Shijie Wu sits in a middle author position without corresponding-author status, consistent with a contributing role within that team rather than an independent line; the collaboration dimension scores 32/100 on this basis.

Research keywords

named entity recognitionin-context learningretrieval-augmented generationlarge language modelslow-resource NLPdemonstration retrievalcross-domain NERentity type confusionlow-resource transferLLM reasoningknowledge augmentation

Highlights

Training-free retrieval reaches low-resource NER state of the art
URA-NER combines retrieval alignment and uncertainty reduction for in-context demonstration selection. Evolve's review reports it lifts Qwen2.5-14B to CrossNER low-resource NER SOTA without task-specific training, with good ablations but no significance tests or leakage controls; repro readiness is rated medium.
Quantifying and repairing entity type confusion in cross-domain NER
This paper introduces MCQM, a multidimensional quantification of NER entity type confusion, and PBRC, LLM-based bidirectional reasoning, reporting strong CrossNER gains. Evolve notes that the SOTA-on-all-domains claim overreaches relative to the paper's own Table 2; repro readiness is rated low.
A tightly scoped low-resource NER agenda
Both 2026 papers attack the same problem from complementary angles: retrieval-augmented demonstration selection and type-confusion reasoning, using LLMs such as Qwen2.5-14B and Flan-T5 in low-resource and cross-domain NER settings. The single-field distribution in natural language processing underpins the 100/100 focus score.

Timeline

2026URA-NER, a unified retrieval-augmented framework with retrieval alignment and uncertainty reduction for low-resource NER, appears on arXiv and enters the Evolve catalog with a medium repro-readiness rating.
2026A second paper quantifies entity type confusion in cross-domain NER via MCQM and adds LLM bidirectional reasoning (PBRC), entering the Evolve catalog with a low repro-readiness rating.
2026With both Evolve-reviewed papers dated this year in the single field of natural language processing, the profile records a 100/100 focus score concentrated on named entity recognition.

Papers in the Evolve catalog

Frequent co-authors

Jingyu Wang×2 Fusheng Jin×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.

Profile is name-resolved from public bibliographic data: ORCID and country are absent from the input and 'Shijie Wu' is a common name, so identity confidence is limited.
Output, impact, and activity dimensions are unscored because publication counts, citations, and per-year activity data are not present in the provided snapshot.
The evidence base is two 2026 arXiv preprints reviewed by Evolve; peer-reviewed venue status is not captured, and no bibliometric totals beyond the listed raw metrics are quoted.
The frequent-coauthor list contains duplicate and self entries (Jingyu Wang twice, Shijie Wu listed); co-author analysis treats it as deduplicated to 4 distinct coauthors.