Shofa, Gevira Zahra (2026) Penerapan model mBART-50 untuk ringkasan abstraktif teks berita bahasa Indonesia dengan evaluasi semantik BLEURT-20. Sarjana thesis, UIN Sunan Gunung Djati Bandung.
|
Text (COVER)
1_cover.pdf Download (330kB) | Preview |
|
|
Text (ABSTRAK)
2_abstrak.pdf Download (125kB) | Preview |
|
|
Text (SK BEBAS PLAGIARISM)
3_skbebasplagiarism.pdf Download (269kB) | Preview |
|
|
Text (DAFTAR ISI)
4_daftarisi.pdf Download (137kB) | Preview |
|
|
Text (BAB I)
5_bab1.pdf Download (299kB) | Preview |
|
|
Text (BAB II)
6_bab2.pdf Restricted to Registered users only Download (711kB) | Request a copy |
||
|
Text (BAB III)
7_bab3.pdf Restricted to Registered users only Download (848kB) | Request a copy |
||
|
Text (BAB IV)
8_bab4.pdf Restricted to Registered users only Download (2MB) | Request a copy |
||
|
Text (BAB V)
9_bab5.pdf Restricted to Registered users only Download (140kB) | Request a copy |
||
|
Text (DAFTAR PUSTAKA)
10_daftarpustaka.pdf Restricted to Registered users only Download (217kB) | Request a copy |
||
|
Text (LAMPIRAN)
11_lampiran.pdf Restricted to Repository staff only Download (1MB) | Request a copy |
Abstract
Indonesia: Perkembangan kegiatan digital seperti mengkonsumsi berita online menyebabkan masyarakat mengalami kesulitan untuk membaca dan memahami inti topik berita yang panjang dalam waktu singkat. Salah satu solusi inovatif yang dapat digunakan yaitu ringkasan teks otomatis abstraktif. Penelitian ini bertujuan untuk mengimplementasikan model mBART-50 dalam menghasilkan ringkasan abstraktif teks berita bahasa Indonesia serta mengevaluasi performa kesamaan semantik hasil ringkasan menggunakan dua metrik otomatis yaitu BLEURT-20 dan BERTScore. Metodologi yang digunakan pada penelitian ini yaitu metodologi SEMMA dengan dataset IndoSum sebanyak 18.774 yang berisi artikel berita bahasa Indonesia. Pada tahap model, model mBART-50 dilakukan fine-tuning menggunakan data latih sebanyak 14.261. Tahap evaluasi menggunakan metrik BLEURT-20 sebagai metrik utama dan BERTScore sebagai metrik pembanding dengan data uji sebanyak 3.762. Hasil penelitian membuktikan bahwa model mBART-50 memiliki performa yang baik dalam menghasilkan ringkasan yang relevan secara semantik. Metrik BLEURT-20 mendapat nilai mean sebesar 0,6997 dan metrik BERTScore F1 mendapat nilai mean sebesar 0,8467. English: The development of digital activities such as consuming online news causes people to experience difficulties in reading and understanding the essence of long news topics in a short time. One innovative solution that can be used is automatic abstractive text summarization. This study aims to implement the mBART-50 model in generating abstractive summaries of Indonesian news texts and evaluate the performance of the semantic similarity of the summary results using two automatic metrics, namely BLEURT-20 and BERTScore. The methodology used in this study is the SEMMA methodology with the IndoSum dataset of 18,774 containing Indonesian news articles. In the modeling stage, the mBART-50 model was fine-tuned using training data of .. The evaluation stage used the BLEURT-20 metric as the main metric and BERTScore as a comparison metric with 3,762 test data. The results of the study prove that the mBART-50 model has good performance in producing semantically relevant summaries. The BLEURT-20 metric gets an average value of 0.6997 and the BERTScore F1 metric gets an average value of 0.8467.
| Item Type: | Thesis (Sarjana) |
|---|---|
| Uncontrolled Keywords: | Ringkasan Teks Abstraktif; mBART-50; BLEURT-20; BERTScore; IndoSum |
| Subjects: | Systems |
| Divisions: | Fakultas Sains dan Teknologi > Program Studi Teknik Informatika |
| Depositing User: | Gevira Zahra Shofa |
| Date Deposited: | 27 Jun 2026 15:13 |
| Last Modified: | 27 Jun 2026 15:13 |
| URI: | https://digilib.uinsgd.ac.id/id/eprint/133341 |
Actions (login required)
![]() |
View Item |



