Penerapan Model Distilbert Dan Pemrosesan Sekuensial Untuk Deteksi Stres Dari Data Teks

Putri, Lia Kharisma (2025) Penerapan Model Distilbert Dan Pemrosesan Sekuensial Untuk Deteksi Stres Dari Data Teks. Other thesis, INSTITUT TEKNOLOGI SEPULUH NOPEMBER.

Text 5025201034-Undergraduate_Thesis.pdf Restricted to Repository staff only until 1 April 2027. Download (6MB) | Request a copy

Abstract

Stres adalah kondisi psikologis yang berdampak negatif pada kesehatan mental. Dengan berkembangnya Natural Language Processing (NLP), deteksi stres berbasis teks kini dapat dilakukan secara otomatis. Penelitian ini bertujuan mengembangkan model deteksi stres menggunakan DistilBERT untuk ekstraksi fitur, yang kemudian dikombinasikan dengan LSTM, GRU, dan RNN untuk menangkap pola temporal dalam teks. Selain itu, penelitian ini mengeksplorasi penggunaan Affect Score, Analisis Sentimen, dan Augmentasi Data untuk meningkatkan akurasi. Dataset yang digunakan adalah Dreaddit, yang berisi teks terkait stres dari komunitas Reddit. DistilBERT menghasilkan representasi teks berbasis embedding, sementara Affect Score dan Analisis Sentimen memperkaya informasi emosional dalam teks. Model LSTM, GRU, dan RNN dilatih dan diuji menggunakan metrik recall dan F1-score. Hasil penelitian menunjukkan bahwa LSTM dengan Affect Score pada epoch 20 memberikan performa terbaik, dengan recall 0,8753 dan F1-score 0,7721. Model ini efektif menangkap hubungan jangka panjang dalam teks dan informasi emosional terkait stres. Sementara itu, RNN dan GRU lebih unggul dalam skenario augmentasi, namun mengalami penurunan performa saat Affect Score ditambahkan tanpa augmentasi. Penelitian ini juga menunjukkan bahwa jumlah epoch berpengaruh terhadap performa model, dengan LSTM optimal pada epoch 20. Penelitian ini menyimpulkan bahwa kombinasi DistilBERT dan LSTM dengan Affect Score adalah pendekatan terbaik untuk deteksi stres berbasis teks, dan dapat digunakan sebagai referensi dalam pengembangan sistem deteksi stres yang lebih akurat di aplikasi seperti analisis media sosial dan intervensi psikologis berbasis AI.
=================================================================================================================================
Stress is a psychological condition that negatively affects mental health. With the development of Natural Language Processing (NLP), text-based stress detection can now be done automatically. This study aims to develop a stress detection model using DistilBERT for feature extraction, which is then combined with LSTM, GRU, and RNN to capture temporal patterns in text. In addition, this study explores the use of Affect Score, Sentiment Analysis, and Data Augmentation to improve accuracy. The dataset used is Dreaddit, which contains stress-related texts from the Reddit community. DistilBERT produces embedding-based representation of text, while Affect Score and Sentiment Analysis enrich emotional information in text. The LSTM, GRU, and RNN models were trained and tested using recall and F1-score metrics. The results showed that LSTM with Affect Score in epoch 20 gave the best performance, with a recall of 0.8753 and an F1-score of 0.7721. This model effectively captures long-term relationships in texts and stress-related emotional information. Meanwhile, RNN and GRU are superior in augmentation scenarios, but experience a decrease in performance when Affect Score is added without augmentation. The study also showed that the number of epochs had an effect on the model's performance, with the optimal LSTM at epoch 20. The study concludes that the combination of DistilBERT and LSTM with Affect Score is the best approach for text-based stress detection, and can be used as a reference in the development of more accurate stress detection systems in applications such as social media analysis and AI-based psychological interventions.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deteksi Stres, DistilBERT, LSTM, GRU, RNN, Affect Score, Analisis Sentimen, NLP, Stress detection, DistilBERT, LSTM, GRU, RNN, Affect Score, Sentiment Analys
Subjects:	T Technology > T Technology (General)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Lia Kharisma Putri
Date Deposited:	06 Feb 2025 07:05
Last Modified:	06 Feb 2025 07:05
URI:	http://repository.its.ac.id/id/eprint/118385

Actions (login required)

View Item