Advancing news headline sarcasm detection through hybrid neural networks

In the modern day, sarcasm post is a regular activity in social media. It is a way of speech in which people use phrases that are not intended to convey their positive or negative feelings. People’s ability to accurately identify sarcasm is declining due to the pervasive usage of this method of communication in headlines on the news and on social media. This study introduces a novel hybrid deep learning model named CBMPBiLSTM (Convolutional Block with MaxPooling and Bi-directional LSTM), designed specifically for sarcasm detection in textual headlines. The model deploys an embedding layer that transforms textual input into dense vector representations, capturing semantic relationships. This is followed by a one-dimensional convolutional layer, which identifies local n-gram features within text. A MaxPooling1D layer is applied to reduce the dimensionality of the feature maps, effectively preserving the most significant and informative features. Further, a Bi-directional LSTM layer enables contextual learning from both past and future word sequences. The network concludes with a dense output layer using sigmoid activation for binary classification. In comparison with other typical hybrid models the proposed model is particularly optimized for news headlines, enabling superior generalization and advancing state-of-the-art performance in sarcasm detection. Experiments conducted on a labeled news headlines dataset demonstrate superior performance, achieving 95.84% accuracy. Comparative results against state-of-the-art methods validate the robustness and effectiveness of the proposed model in identifying sarcastic content.