Natural Language Processing for Author Style Detection

This paper delves into the intricate domain of stylistic features in Natural Language Processing, particularly in the context of long Russian literary texts. Recognizing the prevailing emphasis on content over form in NLP, we assert the importance of textual style in representing literary nuances. Our methodology involves extracting and preprocessing data from an open-source library of Russian literature, spanning a diverse array of authors. We critically survey the field of authorship style detection in literature, delineating its application on short texts, sentences, and whole document embeddings. Opting for the latter, given its suitability for discerning the stylistic signature of extensive works, we introduce a model anchored on the CoLDE architecture. This model employs contrastive learning to discern the unique stylistic imprints of each author. However, challenges in achieving high-quality contrastive perspectives point towards potential areas for further research, such as dataset expansion and batch size adjustments.