Automating Service Classification for Victims of Violence in Indonesia through Deep Learning Using Gated Recurrent Unit and fastText
Dedy Arisandi, Rossy Nurhasanah, Suci Khairiah · 2023
Violence against children and women is a serious issue that is often overlooked. This alarming problem should not be underestimated, as it can lead to severe mental health consequences for the victims. Nonetheless, the existing service provision mechanism continues to rely on manual processes, further hampered by a shortage of personnel relevant for these roles. This combination of factors has given rise to a bottleneck problem, resulting in a slower response to reports of violence. This research introduces a solution based on deep learning approach, using the Gated Recurrent Unit (GRU) and fastText method to automatically identify suitable service based on the information provided in the victim’s report. The dataset comprised 5,799 report sentences, split into training and testing sets. Employing a batch size of 256 and a neuron unit of 128, the model achieved remarkable results with a macro-AUC value of 0.92 and a Hamming loss value of 0.09 during the evaluation process. These outcomes underscore the model’s potential to optimize service provision and accelerate support for victims effectively.