Humor Detection in English-Urdu Code-Mixed Language

Abstract

This research proposes a novel approach for de-tecting humor in code-mixed English-Urdu (Roman Urdu) text. Our approach combines advanced deep learning algorithms, machine learning, and transfer learning algorithms to classify code-mixed text as humorous or non-humorous. We used deep learning algorithms like CNN(Convolutional Neural Networks), LSTM(Long short-term memory), BiLSTM, and a hybrid model made from their combination after some hyper-tuning. We found that the hybrid CNN-BiLSTM model had an accuracy of approximately 75%, while XLM-RoBERTa outperformed all other models with an accuracy of 77.04 %. This is the first time these approaches have been applied to code-mixed Roman Urdu, a low-resource language.