PulseExploreJournal ClubResearchersJournals
Instagram
HomeJournal ClubExplore
Synapse
⌘+K
Synapse
June 20, 2026International Journal of Asian Language Processing

Vietnamese Legal News Classification: A Comparative Study of Transformer-based Models

View Full Paper
Ask AI
Bookmark
Share

Authors

QNQuoc Nguyen

Discussion

Loading...

Member takes

Overview

Randomized trial demonstrates strong accuracy for transformer models in Vietnamese legal news classification, indicating benefits of fine-tuning.

Key Points

  • This research evaluates transformer-based models for classifying Vietnamese legal news articles.
  • Introduced a benchmark dataset of 48,675 legal news articles from two Vietnamese journalism portals.
  • Compared four fine-tuned transformer models: PhoBERT, XLM-RoBERTa, mBERT, and ViELECTRA.
  • Utilized zero-shot and few-shot prompting settings and conducted error analysis including a confusion matrix.
  • PhoBERT and XLM-RoBERTa achieved strong classification performance with 87% accuracy.
  • API-based LLM achieved approximately 60% accuracy but had a refusal rate of over 95% on legal crime-related content.
  • Error analysis revealed Current Events (Thoi su) as a major source of misclassification.

Cite This Study

Quoc Nguyen (2026) studied this question.

synapsesocial.com/papers/6a362ee4db0793dc1a5367cbhttps://doi.org/10.1142/s2717554526500074
View Full Paper
Ask AI
Bookmark
Share