Hybrid-Sarcasm: Sarcasm Detection and Classification Using Hybrid Machine Learning Methods

Abstract

Sarcasm is a nuanced form of linguistic irony in which spoken or written words convey a meaning contrary to their literal sense, making automated detection a persistent challenge in natural language processing (NLP). Existing computational methods struggle because sarcasm relies heavily on contextual cues, prior knowledge, and delivery style. This paper presents Hybrid-Sarcasm, a Hybrid Machine Learning (HML) framework that integrates three distinct feature categories—lexical, sarcastic, and contextual—to classify tweets as sarcastic or non-sarcastic. The proposed approach introduces a sarcasm-specific feature set combined with ensemble classification techniques. Experiments on a Twitter-based dataset demonstrate that the HML classifier achieves 95.30% accuracy on sarcastic feature sets, outperforming baseline methods including K-Nearest Neighbor, Random Forest, Support Vector Machine, and Decision Tree. These results confirm that sarcasm-oriented features substantially improve classifier performance across all models evaluated.