Identifikasi Pernyataan Misogini Berdasarkan Komentar Media Sosial Menggunakan Bidirectional Long Short-Term Memory dan IndoBERT Embedding

Abstract

With the anonymity provided by social media platforms, users are free to upload content or share their opinions through available comment sections. Due to this, it is quite common to encounter accounts in the comment sections that clearly express hatred towards someone, especially towards women, which can be classified as misogyny. Consequently, many women feel distressed and uncomfortable due to receiving comments that belittle, demean, and harass them. Identifying accounts that engage in such behavior is challenging, considering the potential volume of comments and the need for time-consuming manual interpretation of the underlying meaning in comments. In view of this, an approach is needed in system design that has the ability to more effectively identify comments containing misogyny or non-misogyny statements. This study applies a combination of the IndoBERT Embedding method as word embedding and the Bidirectional Long Short-Term Memory algorithm to identify misogyny and non-misogyny comments within social media platforms. The model was developed using 4000 comment samples from social media platforms including Instagram, YouTube, and X. Evaluation of the model using the Confusion Matrix showed an accuracy value of 90%. Considering this, it is feasible to determined that the combination of IndoBERT Embedding and Bidirectional Long Short-Term Memory effectively identifies comments containing misogyny statements as well as those without such statements.