COVID-19 Vaccine News Headline: A Sentiment Analysis with Robustly Optimized BERT Pretraining Approach (RoBERTa)

Abstrak

The COVID-19 situation and its vaccination management necessitate disseminating information, especially news prevalent in Indonesian society. There is a method for classifying information by feelings. This research analyzes the sentiment of COVID-19 vaccine news headlines in Indonesia using the Robustly Optimized BERT Pretraining Approach. The focus is on news headlines from CNN Indonesia, which contain the word "COVID-19 Vaccine." The methods applied include data acquisition by web scraping and preprocessing. Later on, the RoBERTa model was used to classify sentiments under multiclass and binary conditions. The results showed that the sentiment of public opinion regarding the headlines in news releases was favorable rather than negative between January and December 2021. This research was done under two conditions: multiclass and binary. The training-to-testing data ratio was set to 80:20. In the initial scenario, with a multiclass dataset of 3,680 instances, the accuracy is 78.12%, the precision is 78.90%, the recall is 78.12%, and the F1-score is 78.36%. The second scenario, using a binary dataset of 3,466 entries, yielded an accuracy of 81.99%, precision of 84.42%, recall of 86.98%, and an F1-score of 85.68%. This study contributes to comprehending perceptions of the COVID-19 vaccine as portrayed in Indonesian news media. This study demonstrates the efficacy of the RoBERTa model for sentiment analysis, thereby facilitating the construction of precise and efficient NLP-based public opinion monitors. This study confirms the efficacy of the RoBERTa model in sentiment analysis.

Identifier

• DOI: https://doi.org/10.62527/joiv.9.6.3652
• OpenAlex: https://openalex.org/W7117241585