Emotion Detection in Bangla Music Lyrics Using Machine and Deep Learning Approaches

Music has a tremendous impact on human emotions. It may elicit a wide range of emotions, including happiness, enthusiasm, sadness, and rage. This connection between music and emotion has motivated researchers in the field of Music Information Retrieval to create systems for automatic emotion identification in music. The emotional content of a song is conveyed mostly through its lyrics. This research focuses on detecting emotions in Bangla music lyrics using ML and DL techniques. The dataset consists of 4000 Bangla song lyrics, categorized into four emotion classes: Sad, Happy, Romantic, and Relaxation. We employed Bangla Glove for feature extraction and utilized Random Oversampling to address class imbalance. Our study compares the performance of several ML models, including LightGBM, RF, KNN, DT, XGBoost, and SVM. Additionally, we explored DL approaches using ANN, CNN, LSTM, and BiLSTM. To further enhance the performance, we developed hybrid models combining CNN with LSTM, BiLSTM, and GRU, as well as GRU with BiLSTM. Among these models, the RF model achieved the highest accuracy of 0.786, demonstrating its superior capability in capturing complex emotional expressions in Bangla music lyrics. The results present the potential of an ensemble ML in advancing emotion detection from textual data, particularly in the context of Bangla music.