Deteksi Depresi Berbasis Teks Pada Platform X dengan Menggunakan TFIDF-LSTM

Yanuarsyah, Muhammad Jovan Adiiwjaya (2026) Deteksi Depresi Berbasis Teks Pada Platform X dengan Menggunakan TFIDF-LSTM. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5027201025-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Depresi merupakan gangguan kesehatan mental global yang sering kali tidak terdiagnosis akibat stigma sosial dan kurangnya akses ke layanan psikologis. Di Indonesia, media sosial X (sebelumnya Twitter) telah menjadi platform utama bagi pengguna untuk mengekspresikan emosi dan gejala psikologis secara real-time. Namun, deteksi otomatis depresi menghadapi tantangan berupa struktur kalimat informal dan bahasa gaul. Penelitian ini mengusulkan pengembangan sistem deteksi depresi berbasis Deep Learning menggunakan metode Term Frequency-Inverse Document Frequency (TF-IDF) untuk seleksi fitur dan arsitektur Long Short-Term Memory (LSTM). Untuk mengoptimalkan implementasi sistem, penelitian ini juga melakukan analisis komparasi dengan arsitektur Gated Recurrent Unit (GRU) sebagai model pembanding efisiensi komputasi. Model dilatih menggunakan dataset teks tweet yang diklasifikasikan ke dalam tiga kelas: Normal, Moderate, dan Severe. Evaluasi kinerja dilakukan menggunakan Confusion Matrix dengan metrik prioritas F1-Score Macro. Hasil pengujian menunjukkan bahwa seluruh model Deep Learning mampu mengenali indikator depresi dengan sangat baik, di mana model berbasis TFIDF-GRU menghasilkan performa tertinggi dengan Accuracy sebesar 98,67% dan F1-Score sebesar 98,67%. Selain itu, GRU menunjukkan efisiensi komputasi yang unggul dengan waktu pelatihan 2.035,35 detik (sekitar 31,4% lebih cepat dibandingkan Stacked LSTM). Atas dasar efisiensi tersebut, model terbaik berbasis GRU diimplementasikan ke dalam aplikasi web menggunakan framework Flask. Aplikasi web yang dibangun dapat berfungsi sebagai alat skrining awal yang responsif dan efisien bagi pengguna.
===================================================================================================================================
Depression is a global mental health disorder that often remains undiagnosed due to social stigma and limited access to psychological services. In Indonesia, the social media platform X (formerly Twitter) has become a primary venue for users to express emotions and psychological symptoms in real-time. However, automated depression detection faces challenges such as informal sentence structures and slang. This study proposes the development of a Deep Learning-based depression detection system using the Term Frequency-Inverse Document Frequency (TF-IDF) method for feature selection and the Long Short-Term Memory (LSTM) architecture. To optimize system implementation, this study also conducts a comparative analysis with the Gated Recurrent Unit (GRU) architecture as a computational efficiency baseline. The models are trained using a tweet dataset classified into three categories: Normal, Moderate, and Severe. Performance evaluation is conducted using a Confusion Matrix with F1 Score Macro as the priority metric. The experimental results indicate that all Deep Learning models perform exceptionally well, with the TFIDF-GRU model achieving the highest performance, yielding an Accuracy of 98.67% and an F1-Score of 98.67%. Furthermore, GRU demonstrates superior computational efficiency with a training time of 2,035.35 seconds (approximately 31.4% faster than Stacked LSTM). Based on this efficiency, the best performing GRU model is deployed into a web application using the Flask framework. The developed web application serves as a responsive and efficient early screening tool for users.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deep Learning, Deteksi Depresi, Flask, GRU, LSTM, TF-IDF, X (Twitter), Depression Detection
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models. T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Information and Communication Technology > Information Technology > 59201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Jovan Adiwijaya Yanuarsyah
Date Deposited:	04 Aug 2026 01:35
Last Modified:	04 Aug 2026 01:35
URI:	http://repository.its.ac.id/id/eprint/142716

Actions (login required)

View Item