Özet:

Abstract The rapid development of social media has changed the way of interacting and communicating, one of which is using Twitter. Through Twitter, users can express themselves and their feelings directly without limits. It can unconsciously become a medium that reflects one’s personality. In conducting personality assessments, the Natural Language Processing (NLP) model can use to predict personality automatically. So, in this study, an experiment was conducted to predict user personality based on the Big Five Personality Traits, especially in Indonesia. Previous research on personality prediction using BERT has provided promising results. However, BERT has drawbacks because it is limited in processing many words. To process information better it requires prediction of personality at the user-level by using all the user's information.  Based on this, this research focuses on conducting experiments by proposing the Three Order Ensemble method with the BERT workflow (TOEM-BERT) as a scheme for combining tweets so that tweet data can be used optimally. The testing phase consists of two different experimental scenarios using two types of BERT models: IndoBERT and IndoBERTweet. Parallel test scenarios are carried out using the test set for each model, and linear test scenarios are carried out using the same test set for the entire model. The experiments show that the proposed TOEM-BERT method performs better in all test scenarios by obtaining 78.41% Weighted F1 in the linear test using IndoBERT and 77.84% Weighted F1 in the parallel test using IndoBERTweet.

Anahtar Kelimeler: