Objectivity and Subjectivity Classification with BERT for Bahasa Melayu
Wing Kin Chong, Hu Ng, Timothy Tzen Vun Yap, Wooi King Soo, Vik Tor Goh, Dong Theng Cher · 2022
This research present the notion of subjectivity and objectivity in Bahasa Melayu language.Word2Vec and BERT word embedding models are created for the purpose of subjectivity classification and sentiment classification.Two types of embeddings are developed (Word2Vec and BERT) with Wikipedia data as objectivity dataset, Twitter data as subjectivity dataset and combination of both datasets.A pre-trained BERT embedding model called Bert-Base-Bahasa-Cased is used as a reference.First, the datasets are fed into every embedding model to be embedded as vectors.The subjectivity classification and sentiment classification are carried out via 70:30 train-test splits.Both classification tasks are carried out using Logistic Regression, Random Forest, and Double Layer Neural Network classifiers.Logistic Regression on Bert-Base-Bahasa-Cased model achieved the highest result of 99.95% in subjectivity classification and 74.30% in sentiment classification.