Artificial Intelligence in Digital Humanities: Transforming Literary Analysis and Cultural Interpretation

Abstract

The intersection of Artificial Intelligence (AI) with Digital Humanities (DH) is one of the most radically changing tendencies in the modern humanistic studies. In this paper, a detailed exploration of the area of using sophisticated methods of natural language processing (NLP) and machine learning (ML) to analyze literature and interpret it culturally at a large scale are investigated. We present an AI-DH pipeline consisting of multi-layers that combine Latent Dirichlet Allocation (LDA) topic modelling, BERT-based sentiment analysis, transformer-based Named Entity Recognition (NER) and detection of similarity between intertexts on a literature collection of 4 872 English-language texts in the 19th and 20th centuries. In ensemble model, this has a total accuracy of 91.6 percent and a F1-score of 91.1 percent and is significantly more accurate than a single baseline model such as BERT (87.3 percent), RoBERTa (88.9 percent) and traditional SVM classifiers (74.2 percent). Eight thematic clusters are characterized that the most common ones are Romantic Narratives (18.4%) and Political Discourse (14.2). The longitudinal sentiment analysis shows a statistically significant change towards positive affect in the literature after 1960 (r = 0.61, p = 0.001). The analysis also reveals the effectiveness of AI tools to bring out the latent patterns of culture that cannot be identified through conventional methods of close-reading and thus enhance, but not eliminate, humanistic interpretation. They are compared with ten previous studies, which confirm a state of art performance of the proposed architecture. Limitations, moral implications on bias with al algorithms in cultural analytics, and directions to be taken in the future are addressed.