Evaluating the Sentiment Analysis from Auto-Generated Summary Text Using IndoBERT Fine-Tuning Model in Indonesian News Text

Recently, online news has replaced conventional magazines and physical newspapers because of their intuitiveness and timeliness. News sites provide a comprehensive overview of important current events, serving as a valuable source for learning about a country's latest social, political, and economic issues. The government utilizes news channels to get an overview of the specific problems with sentiment analysis. However, the current system only reads news headlines to determine sentiment, so it does not thoroughly measure the opinion in the news content. This situation causes errors in sentiment reading, which should be negatively interpreted as positive or vice versa. This research tests the auto-generated summary text using the IndoBERT fine-tuning model to label the sentiment of news text. This research shows that fine-tuning IndoBERT using the human-made summaries dataset achieves the optimal outcome, with an F1-score of 75% compared to the 65% F1-Score of the auto-generated summary testing dataset. This study shows that the sentiment analysis prediction using the human-made summary dataset scores better than the sentiment analysis resulting from the Autogenerated summary testing dataset.