Analisis Emosi dalam Teks Bahasa Indonesia dengan Pendekatan BERT dan CNN

Abstract

Emotion is an important aspect of human communication and has been widely studied as an important part of human nature. Emotion recognition in text allows for a deeper understanding of the message being conveyed, involving analysis of word choice, punctuation and message length. Emotion recognition in Indonesian texts often involves analyzing adjectives that describe certain emotional states. With the ever-growing amount of information, the challenge of understanding emotions with only manual processing is getting more complex. This research utilizes a combined model of Bidirectional Encoder Representation from Transformer (BERT) and Convolutional Neural Networks (CNN) for text classification. BERT is used to train a language model that can dynamically represent the meaning of words based on their context, and CNN is used to predict the output based on the semantic vector generated by BERT from each word in the text. The classifiable emotions consist of sadness, happiness, love, anger, fear, and surprise. The dataset is taken from the Kaggle site called "Emotions" which contains 300 thousand collections of English Twitter messages. This study obtained 85% accuracy, 92% precision, 86% recall, and 88% F1-score from all emotion classes.