Fine-Tuning IndoBERT Model for Big Five Personality Prediction from Indonesian Social Media

Abstract

The increasing amount of data generated from social media brings opportunities to produce various helpful knowledge and information, one of which is predicting a person’s personality. The advent of attention-based classification techniques and transformers brings promising results on multiple tasks in Natural Language Processing (NLP). In this study, we predicted the Big Five personalities from Indonesian-language social media by fine-tuning the IndoBERT model. The IndoBERT model is a Bidirectional Encoder Representation from Transformers (BERT)-based mono-language model trained in the Indonesian corpus. Based on the experimental results, the prediction model we proposed obtained an accuracy value of 72% and an F1-score of 71% using IndoBERT-base. Meanwhile, using IndoBERT-large can increase the accuracy value by 78% and the F1-score by 74%. The proposed model also outperforms previous models in predicting the Big Five personality.