Explicit Content Classification in Indonesian Song Lyrics Using the LSTM-CNN Method

Music services are currently very popular and accessible to anyone, particularly through internet platforms. However, this widespread accessibility necessitates careful consideration, as music and songs can introduce explicit content to underage children and teenagers, especially through song lyrics. Manual labeling of songs for explicit content is prone to errors and time-consuming, but this process can be improved with Natural Language Processing (NLP) technologies, such as text classification using deep learning. The LSTM-CNN model was chosen for its ability to leverage the strengths of both Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN). LSTM captures long-range dependencies in sequential data, while CNN is effective at extracting localized features from text. This combination makes LSTM-CNN ideal for the complexity of Indonesian song lyrics, which often involve context-dependent content. In this research, the LSTM-CNN model demonstrated optimal performance in classifying explicit content in Indonesian song lyrics. The best results were achieved with a balanced dataset of 5,470 rows, an $80: 20$ data splitting ratio, 5 epochs, a batch size of 128, and a dropout rate of 0.5. The model’s performance metrics were as follows: accuracy of $\mathbf{7 4. 5 2 \%}$, precision of $\mathbf{7 5. 2 2} \%$, recall of $\mathbf{7 4. 3 7 \%}$, and an F1-score of $\mathbf{7 4. 6 8 \%}$.