An Automatic Generation of Poetry using Generative Pre-Trained Transformer with Fine Tuning

In current trends, poetry is a creative expression of thoughts and emotions which manifesting as sequences of words to express ideas in an imaginative style for Automatic generation of poetry. Natural Language Generation (NLG) and generative Artificial Intelligence (AI) generally create a new poem. But the generative AI has lack of poem originality, has limited content which is failed in generating deeper information on the related topic and heavily struggling to create the new concepts with sequential words. To overcome these issues, the model of Generative Pre-trained Transformer 2 with fine tuning (GPT-2-FT) is proposed which is coherent for adapting various forms and contexts to avoid nonsensical sequence of text words. Initially, Poem Generation dataset is used, then in preprocessing, input tokens are divided into different individual words and gives it as input to training words. The proposed GPT-2-FT gives an accurate value and also handles the emotions of poem by Generative Adversarial Network (GAN) which including generator and discriminator. The proposed GPT-2-FT gives improved results in terms of Coherence, accuracy and fluency which values are given by (8.777), (0.998) and (0.975) respectively than existing Bilingual Evaluation Understudy (BLEU) model.