Klasifikasi Cyberbullying Pada Tweet Bahasa Sunda Dengan Menggunakan Hybrid Learning Model

Sari

ABSTRAK
Cyberbullying dalam bahasa Sunda semakin marak di media sosial, dengan kasus seperti penghinaan fisik, body shaming, dan ancaman yang dapat berdampak negatif pada korban. Namun, deteksi otomatis masih menghadapi tantangan, terutama dalam keterbatasan dataset dan efektivitas metode pemrosesan bahasa alami. Penelitian ini bertujuan untuk mengembangkan sistem deteksi cyberbullying bahasa Sunda menggunakan gabungan model stemming dan hybrid learning. Peneliti menerapkan beberapa model machine learning yaitu random forest dan Support Vector Machine (SVM) serta model deep learning yaitu convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM), CNN, dan BiLSTM. Peneliti melakukan eksperimen untuk mengevaluasi kinerja masing-masing model dengan mengukur akurasi dan F1-score. Berdasarkan hasil penelitian, model hybrid learning memperoleh kinerja terbaik dengan akurasi sebesar 97,3% dan F1-score sebesar 97%. Selain itu, waktu pelatihan pada CNN-BiLSTM lebih cepat dibandingkan dengan model lainnya yaitu sekitar 30 detik per epoch.

Kata kunci: Bahasa Sunda, Cyberbullying, Hybrid Learning

ABSTRACT
Cyberbullying in the Sundanese language is becoming more common on social media, with cases like physical insults, body shaming, and threats that can seriously affect victims. However, detecting it automatically remains challenging, mainly due to limited datasets and the difficulty of processing the language effectively. This study aims to develop a Sundanese cyberbullying detection system using a combination of stemming and hybrid learning models. The researchers applied several machine learning models, namely random forest and Support Vector Machine (SVM), and deep learning models, namely convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM), CNN, and BiLSTM. The researchers conducted experiments to evaluate the performance of each model by measuring the accuracy and F1-score. Based on the results, the hybrid learning model achieved the best performance, with an accuracy of 97.3% and an F1-score of 97%. Besides that, the training time on CNN-BiLSTM is faster than the others which is approximately 30 seconds per epoch.

Keywords: Sundanese, Cyberbullying, Hybrid Learning