TOWARDS FIGURATIVE-AWARE SENTIMENT ANALYSIS: A COMPARATIVE REVIEW ON DEEP LEARNING AND TRANSFORMER MODELS

Abstract

Sentiment analysis is a main task in NLP which includes social media analytics, e-commerce, opinion mining, and customer feedback. However, transformer models and deep learning techniques have enhanced contextual understanding; sentiment analysis continues to exhibit endless challenge in figurative language. Sarcasm and metaphor frequently reverse expected sentiment, which causes misclassification in many natural language processing models. Despite this, most studies treat sarcasm and metaphor as separate tasks rather than within a unified framework. This paper presents a comparative review of sentiment analysis techniques, determining the purpose of handling figurative languages across different domains. This analysis highlights the limitations in existing models and enhanced the need for unified frameworks that jointly model sarcasm and metaphor. Index Terms— deep learning, sarcasm detection, metaphor detection, metaphor detection, sentiment analysis, figurative language, natural language processing, transformer models.