Deteksi Berita Persuasif Menggunakan Deep Learning Dengan Optimasi Hyperparameter

Kurniawan, Arya Putra (2025) Deteksi Berita Persuasif Menggunakan Deep Learning Dengan Optimasi Hyperparameter. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025221033-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Beberapa perusahaan menggunakan berita persuasif untuk menarik perhatian konsumen dengan cara yang tidak jujur. Berita persuasif harus disaring untuk memastikan berita yang disajikan objektif. Untuk mengatasi masalah tersebut, penelitian ini mengusulkan pendekatan pembelajaran mendalam untuk mendeteksi berita persuasif berbahasa Inggris menggunakan model Bidirectional Long Short-Term Memory (BiLSTM). Untuk mendapatkan word vector berita persuasif digunakan model penyematan kata Robustly optimized Bidirectional Encoder Representation from Transformer approach (RoBERTa).

Penelitian ini bertujuan untuk membuat metode deteksi berita persuasif dengan bahasa Inggris. Dalam pengujian diperlukan dataset baru untuk mengukur kemampuan model dalam mendeteksi berita persuasif. Dibentuk dataset berita menggunakan bahasa Inggris yang dikumpulkan dari portal berita online Indonesia. Untuk mempermudah pemrosesan data berita baik dalam tahap pelatihan dan pengujian, dilakukan metode text summarization menggunakan algoritma TextRank. Guna memaksimalkan performa model dalam mendeteksi berita persuasif dilakukan fine-tuning dengan penambahan layer Conv1D.

Hasil pengujian menunjukkan bahwa penggunaan fine-tuning pada model pembelajaran mendalam dapat meningkatkan kemampuan mendeteksi berita persuasif hingga 92%. Dibandingkan penelitian sebelumnya yang menggunakan model BiLSTM-RoBERTa, hasil pengujian menunjukkan bahwa model BiLSTM-RoBERTa-Conv1D memiliki keunggulan dalam deteksi berita persuasif.

============================================================
============================================================

Some companies use persuasive news to attract consumer attention in dishonest ways. Persuasive news must be filtered to ensure that the presented news is objective. This study proposes a deep learning approach to detect persuasive English news using the Bidirectional Long Short-Term Memory (BiLSTM) model to address this issue. The Robustly optimized Bidirectional Encoder Representation from Transformer approach (RoBERTa) word embedding model is used to obtain word vectors for persuasive news.

This study aims to create a method for detecting persuasive news in English. A new dataset is required to assess the model’s ability to detect persuasive news. A news dataset in English was formed using content collected from Indonesian online news portals. A text summarization method using the TextRank algorithm was applied to facilitate news data processing during the training and testing stages. To maximize the model’s performance in detecting persuasive news, fine-tuning was performed by adding a Conv1D layer.

The testing results show that fine-tuning the deep learning model can improve the ability to detect persuasive news by up to 92%. Compared to previous research using the BiLSTM-RoBERTa model, the testing results demonstrate that the BiLSTM-RoBERTa-Conv1D model has an advantage in detecting persuasive news.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	berita persuasif, pembelajaran dalam, penyematan kata, hyperparameter, deep learning, persuasive news , text summarization, word embedding
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T59.7 Human-machine systems.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Arya Putra Kurniawan
Date Deposited:	31 Jul 2025 06:09
Last Modified:	31 Jul 2025 06:09
URI:	http://repository.its.ac.id/id/eprint/123713

Actions (login required)

View Item