TRANSFORMER AND BILSTM-BASED SENTIMENT ANALYSIS FOR EMOTIONS IN SHORT STORIES

Abstract

Sentiment analysis of emotions in short stories is essential to determining how readers are affected by the story and how well it communicates its intended themes. However, because narrative texts are complex, current approaches frequently have difficulty adequately capturing the subtleties of mood. In this study, we offer a new model for sentiment analysis in short tales, based on Transformer with attention concept and BiLSTM. By exploiting the beneficial qualities of Transformer architecture for capturing global dependencies and BiLSTM for collecting local context, this model tackles the drawbacks of older methods. Combining these two architectures allows our model to provide an extensive analysis of sentiment variance while accurately capturing the story's emotional direction. A considerable improvement over previous methods was achieved with a valence score of around 76% and an arousal score of around 62% when our model's performance was assessed using common metrics like arousal and valence percentages. Our study offers insights into the emotional landscapes of short tales that may impact a variety of sectors, including education, movie directors, other script writers and mental health. It goes beyond academic research to contain significant practical consequences for society. For example, content producers might modify their narratives to elicit particular emotional reactions in their audience, and educators can use our approach to assess the emotional impact of instructional stories on students. The proposed Transformer and BiLSTM-based sentiment analysis methodology opens up new avenues for study on sentiment-aware narrative development and analysis, paving the way for future research in sentiment-aware narrative generation and analysis.