Implementasi Algoritma Gated Recurrent Unit untuk Identifikasi Provokasi Doxing di Media Sosial X

Abstract

The rapid development of social media has transformed the digital communication landscape, but it has also given rise to various forms of abuse, such as provocative doxing. This phenomenon is a planned act of provocation that encourages the dissemination of personal information without consent, thus threatening the privacy and security of users. This research aims to develop an automatic identification system to detect provocative content that has the potential to trigger doxing activities on the X Social Media platform. The research methodology uses a machine learning approach by implementing the Gated Recurrent Unit (GRU) algorithm integrated with the FastText word embedding technique. The combination of these two methods was chosen based on their ability to process sequential data and understand the semantic context of language. The research dataset was acquired by crawling 3050 tweets from Social Media X, which then went through preprocessing and labeling stages to ensure the quality of the model training input. The experimental results show that the GRU model with FastText embedding achieves an accuracy rate of 87.21% based on confusion matrix evaluation. Precision, recall, and F1-score metrics substantiate this performance, demonstrating the model's ability to distinguish provocative content from regular content. The significance of this research lies in its potential implementation as an automated filtration mechanism that can be integrated into social media messaging systems, especially anonymous messaging features to mitigate the risk of spreading sensitive information.