Automated Sarcasm Detection in English Tweets Using CCNN and ELLSTM with Text and Emoji Embeddings
Sinan Ibrahim, Manoj M. Deshpande, Vijaykumar N. Pawar · 2025
Automatic detection of sarcasm is one of the most challenging tasks in natural language processing; hence, extending it into the realm of English tweets-where sarcasm is usually conveyed by text and emojis-would add to the complexity. The paper addresses a new avenue in using a Customized Convolutional Neural Network combined (CCNN) with Extreme Learning Long Short-Term Memory (ELLSTM) for sarcasm detection in tweet data in English. Sarcasm is particularly challenging due to the scarcity of annotated corpora and the subtleties inherent in its detection, which relies mostly on context incongruity. This model will draw upon both linguistic and visual cues-text in tweets, along with accompanying emojis serving as non-verbal contextual indicators, akin to facial expressions-to improve the accuracy of detection. Emoji embeddings provide additional contextual information that has been identified as important for the detection of sarcasm. Experiments have been conducted with the CCNN-ELLSTM model using a dataset of sarcasm-annotated English tweets. Preliminary results demonstrate the effectiveness of the proposed model in achieving the best accuracy and F-score value of 97.57% and 0.9718, respectively. The effectiveness of text-emoji anchoring combos is shown in this work, which also supports the idea that automatic feature extraction might provide reliable and effective algorithms for recognizing sarcasm in English tweets.