IPB University Logo

SCIENTIFIC REPOSITORY

IPB University Scientific Repository collects, disseminates, and provides persistent and reliable access to the research and scholarship of faculty, staff, and students at IPB University

AI Repository
 
Building and Categories


      View Item 
      •   IPB Repository
      • Final Assignments
      • Undergraduate Final Assignments
      • UF - School of Data Science, Mathematic and Informatics
      • UF - Statistics and Data Sciences
      • View Item
      •   IPB Repository
      • Final Assignments
      • Undergraduate Final Assignments
      • UF - School of Data Science, Mathematic and Informatics
      • UF - Statistics and Data Sciences
      • View Item
      JavaScript is disabled for your browser. Some features of this site may not work without it.

      Pemodelan Topik Berita Online Menggunakan BERTopic dengan Sentence Embedding IndoSBERT dan Multilingual MPNet

      Thumbnail
      View/Open
      Cover (414.9Kb)
      Fulltext (2.004Mb)
      Lampiran (609.9Kb)
      Date
      2026
      Author
      Indriyani, Cindy
      Anisa, Rahma
      Mualifah, Laily Nissa Atul
      Metadata
      Show full item record
      Abstract
      Perkembangan media digital menghasilkan volume berita online besar dan tidak terstruktur, sehingga memerlukan pemetaan isu otomatis guna menangkap arah kebijakan publik. Metode yang relevan adalah BERTopic yang mengandalkan model embedding sebagai komponen penentu representasi makna dokumen. Oleh karena itu, penelitian ini membandingkan model sentence embedding IndoSBERT yang mampu menangkap karakteristik kosakata lokal dengan multilingual MPNet sebagai model lintas bahasa. Tujuan penelitian ini mencakup analisis karakteristik embedding IndoSBERT dan multilingual MPNet, evaluasi pengaruh optimasi hyperparameter berbasis Optuna pada model BERTopic, serta identifikasi topik dominan dan dinamika temporal pemberitaan. Penelitian ini menggunakan detikcom sebagai sumber data representatif dinamika isu di masyarakat. Data yang dianalisis mencakup 169.446 berita periode Januari 2023 hingga Desember 2025. Dataset diproses menggunakan BERTopic dengan integrasi UMAP, K-Means++, dan hierarchical clustering sebagai alat interpretasi. Hasil analisis menunjukkan bahwa IndoSBERT lebih baik dalam menangkap konteks kalimat tunggal berbahasa Indonesia secara mendalam karena efisiensi proses tokenisasi serta keunggulan arsitektur modelnya. Melalui optimasi Optuna, BERTopic-IndoSBERT sebagai model terbaik mencapai skor topic quality (TQ) sebesar 0,5548 dengan 38 topik. Melalui pendekatan hierarki, topik tersebut terorganisasi ke dalam sembilan klaster dengan dominasi klaster Kebijakan Publik dan Tata Kelola Pemerintahan (24,15%). Analisis ini menangkap pergeseran fokus pemberitaan dari kompetisi politik menuju implementasi kebijakan pemerintah.
       
      The development of digital media has resulted in a large and unstructured volume of online news, necessitating automated issue mapping to capture public policy directions. A relevant method is BERTopic, which relies on an embedding model as a determining component in document meaning representation. Therefore, this study compared the IndoSBERT sentence embedding model, which is capable of capturing local vocabulary characteristics, with the multilingual MPNet as a cross-language model. The objectives of this study included analyzing the characteristics of IndoSBERT and multilingual MPNet embeddings, evaluating the effect of Optuna-based hyperparameter optimization on the BERTopic model, and identifying dominant topics and the temporal dynamics of news coverage. This study used detikcom as a representative data source of issue dynamics in society. The analyzed data included 169.446 news items from January 2023 to December 2025. The dataset was processed using BERTopic with the integration of UMAP, K-Means++, and hierarchical clustering as interpretation tools. The analysis results showed that IndoSBERT was better at capturing the context of single Indonesian sentences in depth due to the efficiency of the tokenization process and the superiority of its model architecture. Through Optuna optimization, BERTopicIndoSBERT, as the best model, achieved a topic quality (TQ) score of 0,5548 with 38 topics. Using a hierarchical approach, these topics were organized into nine clusters, dominated by the Public Policy and Governance cluster (24,15%). This analysis captured the shift in news coverage focus from political competition to government policy implementation.
       
      URI
      http://repository.ipb.ac.id/handle/123456789/173438
      Collections
      • UF - Statistics and Data Sciences [103]

      Copyright © 2020 Library of IPB University
      All rights reserved
      Contact Us | Send Feedback
      Indonesia DSpace Group 
      IPB University Scientific Repository
      UIN Syarif Hidayatullah Institutional Repository
      Universitas Jember Digital Repository
        

       

      Browse

      All of IPB RepositoryCollectionsBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

      My Account

      Login

      Application

      google store

      Copyright © 2020 Library of IPB University
      All rights reserved
      Contact Us | Send Feedback
      Indonesia DSpace Group 
      IPB University Scientific Repository
      UIN Syarif Hidayatullah Institutional Repository
      Universitas Jember Digital Repository