Evaluating the Effectiveness of AI-Assisted Emotional Metadata in Enhancing the Discoverability of Literary Texts in Digital Libraries

Abstract

The paper is an examination of how AI-assisted emotional metadata can be used to enhance discoverability of digital library texts. Conventional library metadata is mostly based on factual and descriptive labels and this constrains people in their search as they cannot seek emotional or affective features. The study follows the mixed methodology, using both the quantitative analysis of the AI based emotional tags with the qualitative data of the human specialists, librarians and literature scholars. Our example is the review of two W. H. Davies poems, the Rain and Leisure with the help of AI models (GPT, BERT, RoBERTa), and the results were evaluated compared to human-assigned emotion metadata in terms of accuracy, match percentage, and depth of interpretation. It has been found that AI is able to produce accurate emotional signals on emotions fairly superficially (calmness and serenity), but it has difficulty with subtler aspects of literature, thus the need for human intervention. The research shows that discovery in digital libraries, user interaction, and search can be boosted by adding emotional metadata in digital libraries. Through a hybrid strategy of maximum efficiency of AI and human interpretive skills, libraries will be able to update the current cataloguing methods and save the literary and cultural integrity of their material. The results are useful in library metadata standards and give guidelines of future studies in AI-aid to digital humanities.