Exploring Abstractive Indonesian News Text Summarization Using BART Model

Abstractive text summarization is one of the most important tasks in the Natural Language Processing (NLP) field. It involves generating concise summaries while also preserving the original text’s meaning. However, implementing abstractive summarization for Indonesian language presents challenges due to the lack of large-scale datasets and pre-trained model. While many studies have explored various models, the Bidirectional and Autoregressive Transformer (BART) architecture remains underexplored. This study implements abstractive summarization for the Indonesian language using fine-tuned BART model. The dataset used for this study is the IndoSum dataset, which consists of Indonesian news articles. The BART tokenizer was implemented for the preprocessing to fit the model input constraints. The model was evaluated using Recall-Oriented Understudy for Gisting Evaluation (ROUGE) metrics. The experimental result shows that the BART model achieved ROUGE-1, ROUGE-2, and ROUGE-L of 71.62, 64.44, and 69.19 respectively, outperformed T5 and GPT-2. These findings suggest how well BART can capture semantic coherent and produce logical summaries, indicating effectiveness for abstractive Indonesian news text summarization task.