Core Concepts
Introducing AraPoemBERT, a pretrained language model exclusively for Arabic poetry analysis, achieving state-of-the-art results in various NLP tasks related to Arabic poetry.
Abstract
The content introduces AraPoemBERT, a BERT-based language model pretrained on Arabic poetry text. It outperforms other models in tasks like poet's gender classification and poetry sub-meter classification. The dataset used contains over 2.09 million verses associated with attributes like meter, sub-meter, poet, rhyme, and topic. AraPoemBERT demonstrates effectiveness in understanding and analyzing Arabic poetry.
Directory:
Introduction to Arabic Poetry Analysis
Classical Meters in Arabic Poetry
Non-Classical Meters in Arabic Poetry
Transformers in NLP
Proposed Model: AraPoemBERT
Experiments and Results
Stats
AraPoemBERT achieved unprecedented accuracy in poet’s gender classification (99.34% accuracy).
The model achieved an accuracy score of 97.73% in poems’ rhyme classification.
AraPoemBERT significantly outperformed previous works and other comparative models.