Using Pre-Trained Models for Sentiment Analysis in Indonesian Tweets

Sentiment analysis from Twitter can provide useful information about public opinion to public perceptions of a topic, product, and service provided by industry players to government. This information can be useful to be the basis for consideration in making strategies, policies, and decisions making. One of the approach methods used for sentiment analysis is supervised learning. However, the construction of the scratch form classification model requires considerable effort and time. Pre-trained model libraries for sentiment analysis namely Vader, Textblob, and Flair can be used to reduce this burden. In addition, this pre-trained model can be used for the process of labeling training data in the supervised learning method approach. In this study, the challenges of sentiment prediction by the three pre-trained models were discussed on Indonesian text data. As a result, pre-trained models outperform classifier built from scratch. In the comparison of the accuracy levels of the three pre-trained models, the sentiment predictions made by Vader were more accurate than those of Textblob and Flair.