Detection of Depression on Social Media X With FastText Feature Expansion Using Hybrid Deep Learning CNN-GRU

Depression is a mental illness that is experienced by many people in the world. Depression can have serious repercussions such as suicide if not treated early. Therefore, it is important to conduct early detection to provide mental support and appropriate treatment. Detection can be done through one way, namely identifying posts on social media. Thus, this study aims to detect depression through Indonesian tweets on social media X using a hybrid deep learning approach, namely Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), CNN+GRU, and GRU+CNN, by utilizing FastText feature expansion and applying optimization. The dataset used consists of 50,523 data and corpus similarity with a total of 100,594 used in feature expansion. This research evaluates the model performance using five scenarios including Split Data, NGram, Max Features, Feature Expansion, and Optimization. The best results were obtained with Split Data 90:10, Unigram + Bigram + Trigram, max features 10,000 for CNN and GRU, and 5,000 for the hybrid model. Feature expansion using FastText Top 1 on the combination of Tweet + IndoNews dataset improves hybrid model performance. The Nadam optimizer provides optimal results for CNN+GRU, while Adam is optimal for GRU+CNN. The CNN+GRU and GRU+CNN hybrid models achieved the best accuracy of 83.19% with an improvement of 1.36% and 83.32% with an improvement of 1.44%, respectively, showing significant improvement over the baseline.