Leveraging LLM-Generated Lyrics Synthetic Dataset on Indonesian Song Lyrics Age Classification

The popularity of Indonesian songs in social media and streaming services has caused concern to emerge over age-inappropriate lyrics, particularly their potential impact on the behavior and emotions of children and adolescents. Therefore, there is a need to develop a model capable of classifying Indonesian song lyrics into four age groups of listeners: children, adolescents, adults, and all ages. The latest development of age-based Indonesian Lyric Identification employs the Naïve Bayes Classifier (NBC) algorithm, resulting in mediocre results of 65 % accuracy with 80 % training data ratio and 20 % testing data ratio. In this paper, we propose a novel approach incorporating a Large Language Model (LLM), specifically the LLAMA3 70 billion parameters, to enhance classification performance and augment the number of datasets by up to 1000 %. By expanding the dataset tenfold, we improved the performance significantly using hyperparameter tuning on IndoBERT multiclass classification.