Evaluating AI-Assisted Methods for Large-Scale Literary Text Analysis

Abstract

The rapid expansion of digitized literary archives has transformed the study of literature from a predominantly close reading discipline into a data-intensive field supported by artificial intelligence and computational linguistics. This research evaluates AI-assisted methods for large-scale literary text analysis and investigates their validity, interpretability, and contribution to literary scholarship. While machine learning, topic modeling, and natural language processing enable the examination of millions of pages, scholars continue to debate whether these techniques capture the nuanced aesthetic and cultural dimensions traditionally explored through human interpretation. The study develops a framework that compares three families of AI methods: semantic modeling, stylistic pattern detection, and narrative structure extraction. The objective is to determine how effectively these approaches support thematic discovery, authorship attribution, and genre classification. A mixed methodological design was employed using a corpus of 12,000 English novels from the nineteenth and twentieth centuries. Quantitative indicators of algorithmic performance were combined with expert evaluations from literary scholars. SmartPLS structural modeling was used to validate relationships between methodological transparency, interpretive usefulness, and scholarly acceptance. Results indicate that AI methods achieve high accuracy in pattern recognition tasks but vary considerably in their capacity to support meaningful interpretation. Semantic models based on transformer architectures demonstrated superior thematic coherence, while traditional topic models showed limitations in representing metaphor and irony. The research further reveals that interpretability mediates the relationship between technical performance and scholarly adoption. The study contributes to digital humanities by offering an evaluative index for AI-assisted literary analysis that balances computational efficiency with hermeneutic value. Practical implications include guidelines for corpus preparation, model selection, and collaborative workflows between data scientists and literary experts. The findings suggest that AI should be treated as an exploratory partner rather than a replacement for critical reading. Future research should extend the framework to multilingual corpora and investigate ethical concerns surrounding algorithmic bias and copyright. By grounding technological enthusiasm in methodological reflection, the study supports a more thoughtful integration of artificial intelligence into literary scholarship.