A Novel End-to-End Framework for Story Generation Using Deep Neural Networks

Ayan Sar, Purvika Joshi, Subhangi Sati, Richa Choudhary, Sumit Aich, Tanupriya Choudhury, Ketan Kotecha, Turgut Ozseven

Abstract: This paper introduces a novel end-to-end framework for story generation utilising deep neural networks. Generating text belongs to a vital part of Natural Language Processing (henceforth, NLP), and different areas of its applications, including creative writing, entertainment, and education, made it a subject of this research interest. Traditional approaches often involve multi-step pipelines, leading to challenges in coherence and creativity. Our method, in contrast, makes use of deep neural network techniques to craft one story in one single process where all the building blocks connect in a coherent narrative. We provide an overview of deep learning techniques in NLP and highlight the advantages of end-to-end learning approaches. The architecture harnesses the latest available technology for neural network design with a primary focus on generative story writing. We discuss the importance of high-quality datasets and efficient preprocessing techniques to train robust models. Findings have been complimented by methods of evaluation metrics aimed at assessing story quality and coherency. Through case studies and applications, we demonstrate the effectiveness of our framework in various domains, including creative industries and education. Finally, we discuss future directions and challenges, outlining opportunities for advancing the field of story generation using deep neural networks.

External IDs:doi:10.1145/3660853.3660923