A study on the efficacy of natural language generation techniques for similar writing personalities

Abstract

Over the past two decades, the domain of Natural Language Processing has undergone a remarkable transformation enabling machines to generate text, summarize content, paraphrase and analyze sentiment. The captivating idea of analyzing and copying someone’s writing style is no longer impossible. Pushing boundaries further, we have embarked on a journey to implement such a model for the Bengali language by utilizing the approach of style transfer through the application of deep learning using LLM. One’s writing personality can be identified by training the model by imputing a set of documents (notes, books, etc) authored by the writers only. The system will be able to extract important information to recreate sentences with similar structural properties used by the author. Additionally, it will also be able to detect whether a particular sentence structure synchronizes with that author’s distinctive style. The outcome of our model aims to fulfill the need of a particular writing taste of an author, as requested by the user. In essence, our model blends technology and art to write in a way that is reminiscent of their favorite Bengali author. Our proposed model not only skillfully excels in authorship classification and mimicking their style, but also stands resilient against potential adversarial attacks, making it a strong and unyielding system that aligns well with our research objective.