Klasifikasi Cyberbullying pada Media Sosial Komentar Instagram Bahasa Indonesia Menggunakan Metode Bidirectional Long-Short Term Memory (BiLSTM)

Abstract

Indonesia is one of the countries with the highest number of social media users in the world. According to data from katadata.co.id, the number of social media users in Indonesia reached 96 million in 2017 and is estimated to increase by more than 25.3 million by 2022, reflecting an increase of over 30%. A survey conducted by the Indonesian Internet Service Providers Association (APJII) revealed that approximately 49% of internet users have been victims of cyberbullying on social media. Additionally, the Indonesian Child Protection Commission (KPAI) receives hundreds of reports of cyberbullying each year. A GlobalWebIndex study showed that 61% of respondents use social media during their leisure time. This study aims to identify words or combinations of words with the highest potential for cyberbullying on social media, particularly Instagram. The dataset used in this study consists of 8000 Indonesian-language comments collected through the Instagram application using the instaloader library. The classification method employed is the Bidirectional Long Short Term Memory (BiLSTM) algorithm. The results indicate that the BiLSTM algorithm can classify cyberbullying comments with an accuracy rate of 95.5%. The model divides the training and testing data with a proportion of 80% and 20%, respectively, and can classify comments into four categories: sexism, flaming, body shaming, and neutral. Based on these results, the BiLSTM model demonstrates good performance and reliability for classifying cyberbullying comments on Indonesian-language Instagram.