Reproduce “Arabic Tweet Act: A Weighted Ensemble Pre-Trained Transformer Model for Classifying Arabic Speech Acts on Twitter” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Khadejaa Alshehri, Areej Alhothali, Nahed Alowidi

Researchers from King Abdulaziz University developed a weighted ensemble of pre-trained transformer models to classify six categories of Arabic speech acts in dialectal tweets. Their approach, supported by a new 22,352-tweet dataset and data augmentation, achieved an accuracy of 0.85 and a macro-averaged F1 score of 0.74 on the new dataset, outperforming baseline models and improving minority class performance.

Key insights

  • Transformer-based models, especially an ensemble approach combining tweet-specific BERT variants, consistently outperform traditional deep learning baselines for Arabic speech act classification.
  • Data augmentation significantly enhances model performance on minority classes, leading to substantial improvements in macro-averaged F1 scores across diverse Arabic tweet act datasets.
  • LIME analysis revealed that misclassifications often stem from inherent ambiguity and overlap of speech act intentions in short, informal dialectal Arabic social media texts.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.