A Method for Classifying Sarcasms Based on Quantum TF_IDF Features

This article proposes an LSTM-based approach for sarcasm detection by forming quantum features from textual data. In the proposed method, words are first encoded in quantum space, and their semantic similarity is calculated using interference. Based on the encoded data, quantum TF-IDF features are generated and fed into a neural network for classification. The obtained results were compared with those of the classical approach. According to experimental results, the quantum approach achieved 83% accuracy, while the classical approach achieved 78%. Additionally, the quantum approach reduced computation time by 49% compared to the classical one. Due to its superposition and parallelism properties, the quantum approach provides higher performance. Its limitations include testing on a simulator, interdependence between data, and the inability to use real quantum technologies. Applying quantum computing through TF-IDF methods for natural language processing and sarcasm detection can be a promising direction.