Home / Current Issue / Paper 1713456
Transformer-Based Emotional State Classification in Poetic Texts
Subject area: Science,Engineering and Technology · Area of research: Natural Language Processing
DOI: https://doi.org/10.64388/IREV9I7-1713456
Abstract
While emotion recognition is a core task of NLP, its application to poetry is hindered by the heavy use of metaphors and irregular structures that bypass simple keyword-based detection. This paper details a deep learning approach using a fine-tuned BERT encoder to classify emotions in English poetry across nine categories, including Love, Joy, Courage, and Sadness. Our methodology utilizes deep contextual embeddings and standard regularization to ensure high performance and model stability. Empirical testing demonstrates the model?s effectiveness, reaching an accuracy of 89.75% and a weighted F1-score of 89.72%. The findings suggest that transformer models are exceptionally well-suited for decoding the complex affective signals found in literature. We conclude by addressing limitations regarding overlapping emotional states and proposing next steps for intensity-aware and cross-lingual emotion detection.
Keywords
Poetry Emotion Classification, BERT, Transformer Models, Semantic Embeddings, Deep Learning, Natural Language Processing, Affective Computing.
How to cite this paper
@article{1713456,
author = {Ayush S, Harshith Raj Gowda H S, Vinay M G},
title = {Transformer-Based Emotional State Classification in Poetic Texts},
journal = {Iconic Research And Engineering Journals},
year = {2026},
volume = {9},
number = {7},
pages = {438-444},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1713456.pdf},
abstract = {While emotion recognition is a core task of NLP, its application to poetry is hindered by the heavy use of metaphors and irregular structures that bypass simple keyword-based detection. This paper details a deep learning approach using a fine-tuned BERT encoder to classify emotions in English poetry across nine categories, including Love, Joy, Courage, and Sadness. Our methodology utilizes deep contextual embeddings and standard regularization to ensure high performance and model stability. Empirical testing demonstrates the model?s effectiveness, reaching an accuracy of 89.75% and a weighted F1-score of 89.72%. The findings suggest that transformer models are exceptionally well-suited for decoding the complex affective signals found in literature. We conclude by addressing limitations regarding overlapping emotional states and proposing next steps for intensity-aware and cross-lingual emotion detection.},
keywords = {Poetry Emotion Classification, BERT, Transformer Models, Semantic Embeddings, Deep Learning, Natural Language Processing, Affective Computing.},
month = {January},
doi = {https://doi.org/10.64388/IREV9I7-1713456}
}