PAP_NER: A large-scale vietnamese administrative named entity recognition corpus and hybrid deep learning architecture

Clinical Snapshot

Key Findings

• P Value: p < 0.05 (McNemar's test) for CRF layer contribution on Legal Document (VBPL) entity type in ablation analysis

• Effect Size: Performance improvement over BiLSTM+CRF: +2.01% Micro F1; over multilingual XLM-RoBERTa: +2.52% Micro F1; over pure Transformer (PhoBERT without CRF): +0.44% Micro F1

• Primary Outcome: PhoBERT-CRF achieves 97.95% Micro F1-score on the PAP_NER test set for Vietnamese administrative Named Entity Recognition across five entity types

• Nnt Or Sensitivity: Not applicable in traditional clinical sense. Inter-annotator agreement κ = 0.85 (annotation reliability). CRF layer improvement for VBPL entities: +0.96% F1 (p < 0.05). No sensitivity/specificity breakdown by entity type reported.

• Confidence Interval: Not reported — a significant methodological omission

Clinical Application

The PAP_NER corpus and PhoBERT-CRF architecture are publicly available and technically implementable by NLP teams with standard GPU infrastructure. Integration into Vietnamese government document management systems is feasible given the high F1 performance, though production deployment would require domain-specific validation and human oversight protocols. Computational requirements for PhoBERT inference are moderate but not trivial for resource-constrained settings. Direct clinical applicability to Australian healthcare practice is minimal. However, indirect relevance exists in several domains: (1) Australian digital health informatics teams working with multilingual patient populations generating Vietnamese-language administrative documents (e.g., My Health Record, Medicare correspondence for Vietnamese-speaking patients); (2) The methodological framework for corpus construction and NER benchmarking is directly applicable to Australian clinical NLP initiatives (e.g., Australian Clinical NLP, SNOMED CT-AU entity recognition); (3) The human-in-the-loop annotation methodology aligns with RACGP and ADHA standards for clinical data quality; (4) PBS and TGA document processing automation could benefit from analogous NER approaches applied to English-language regulatory and administrative text. No PBS listing, TGA registration, or RACGP guideline is directly relevant to this NLP resource paper. This study has no direct clinical patient population. It is applicable to Vietnamese government agencies, e-Government system developers, public health informatics teams processing Vietnamese administrative health documents, and NLP researchers working on low-resource language administrative text processing

Abstract

Named Entity Recognition (NER) is fundamental for automating administrative document processing in digital government systems. However, Vietnamese NLP research faces a critical infrastructure gap: existing datasets focus on generic information extraction (news, medical) rather than domain-specific administrative text. We present PAP_NER, the first large-scale, gold-standard Vietnamese administrative NER corpus comprising 162,801 sentences with 205,807 entity annotations across five entity types critical for e-Government workflows: Agency (CQ), Legal Document (VBPL), Object (ĐT), Datetime (NG), and Quantity (SL). The dataset was constructed through a rigorous human-in-the-loop annotation pipeline, achieving an inter-annotator agreement of κ = 0.85. We demonstrate PAP_NER's value through comprehensive benchmarking of an established hybrid deep learning architecture, PhoBERT-CRF, which couples monolingual Transformer embeddings (PhoBERT) with Conditional Random Fields for structured prediction. PhoBERT-CRF achieves 97.95% Micro F1-score on the PAP_NER test set, significantly outperforming established baselines: BiLSTM+CRF (+2.01%), multilingual XLM-RoBERTa (+2.52%), and pure Transformer approaches (+0.44%). Ablation analysis reveals that the CRF layer provides statistically significant improvements for structurally complex entities (VBPL: + 0.96%, p < 0.05, McNemar's test). We release PAP_NER publicly (DOI: 10.5281/zenodo.18044019) under Creative Commons BY 4.0 license to support reproducibility and enable further research in Vietnamese administrative NLP. This work establishes a foundational dataset and methodology for addressing the Vietnamese government NER gap, with implications for low-resource language NLP research.

References

1. 1.La, D.-D., Tran, T.-B., Du, N.-H., Dang, N.-H., Phung, T.-N., & Tran, V.-K. (2026). PAP_NER: A large-scale vietnamese administrative named entity recognition corpus and hybrid deep learning architecture. PLOS ONE. https://doi.org/10.1371/journal.pone.0353166