Moodly

This study utilizes three datasets for sentiment analysis: Sentiment140 (1.6 million tweets labeled as positive, negative, or neutral), the IMDb Dataset (50,000 movie reviews for binary sentiment classification), and the Hugging Face Multiclass Sentiment Analysis Dataset (sentences labeled as positive, negative, or neutral). The data were preprocessed through text normalization, removal of URLs, mentions, hashtags, punctuation, repeated characters, and emojis, as well as tokenization, stopword removal, and lemmatization using NLTK. For encoding, the BERT model was employed to generate contextual sentence embeddings ([CLS] vectors), which were then fed into quantum models. Dimensionality reduction was performed using PCA for QSVM and Quantum Random Forest, while quantum neural networks included a trainable linear layer to dynamically project vectors into the space corresponding to the number of qubits. The system interface is built with Flutter, providing a cross-platform mobile and web user experience, while the backend API is developed in Django, handling data requests, model inference, and communication between the frontend and the quantum classification models efficiently.