Poetry Form Identification - A Comparative Analysis of Embedding Models and Classifiers

In recent years, automated text classification has advanced significantly, yet poetry classification remains a relatively under-explored domain. By employing Machine Learning and pre-trained language models, the proposed paper aims to automate poetry classification and provide a comparative analysis for utilizing GPT2 (Generative Pre-Trained Transformer), BERT (Bidirectional Encoder Representations from Transformers), and RoBERTa (Robustly Optimized BERT Pre-Training Approach). We have employed tokenization, sequence padding, and fine-tuning on a diverse dataset of poetry forms, including Sonnet, Haiku, Limerick, Acrostic, Ballad, Epigram, Sestina, and Villanelle. The proposed model is trained and evaluated on a dedicated data set using the application of AdamW Optimizer, and strives for robust classification performance. This research study yields promising results, with the Support Vector Machine (SVM) classifier achieving the highest accuracy of 93.25 percent using the top 20 percent ranked features. This study contributes to the field of automated poetry analysis by integrating various embedding models and classifiers, presenting notable advancements in the categorization of poetry.