Performance Analysis of ChatGPT for Indonesian Abstractive Text Summarization

Effective text summarization is becoming more and more important for enabling rapid understanding and analysis of massive amounts of data. This work aims to develop text that is both brief and resembles human-like coherence by using prompt tweaking and few-shot learning techniques to build summaries from two Indonesian datasets: Liputan6 and Indosum. By using this method, the study shows that sophisticated summary algorithms can produce concise and understandable summaries. These summaries are evaluated using both human assessment and automatic metrics, with an emphasis on the ROUGE score, in order to determine how effective the created content is. Differences were found between the ROUGE scores and human preferences, suggesting that traditional machine evaluation metrics in text summarization might not entirely match human evaluative norms. This discrepancy raises the question of whether assessment techniques should be improved in order to more accurately reflect the subtleties of human judgment. The results point to a possible direction for improving the efficacy and evaluative precision of automated summarization systems in a range of contexts.