Transliterasi Aksara Jawa Berbasis Sequence-to-Sequece Menggunakan LSTM Dengan Mekanisme Attention

Norfaize, Ike (2026) Transliterasi Aksara Jawa Berbasis Sequence-to-Sequece Menggunakan LSTM Dengan Mekanisme Attention. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221199-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Aksara Jawa merupakan salah satu warisan budaya Nusantara yang banyak digunakan dalam berbagai manuskrip kuno. Meskipun penelitian terdahulu telah berhasil mendeteksi karakter aksara Jawa dari citra manuskrip, hasilnya masih berupa urutan kelas aksara yang sulit dipahami oleh orang awam. Transliterasi aksara Jawa ke huruf Latin menjadi tantangan tersendiri karena aksara Jawa bersifat abugida dan memiliki hubungan antar-aksara yang kompleks. Oleh karena itu, penelitian ini mengembangkan sistem transliterasi aksara Jawa berbasis metode Sequence-to-Sequence (Seq2Seq) menggunakan Long Short-Term Memory (LSTM) dengan mekanisme Bahdanau Attention. Dataset yang digunakan berasal dari penelitian sebelumnya yang menyertakan pasangan urutan kelas aksara Jawa dan teks Latin.
Tahapan metodologi dalam penelitian ini meliputi prapemrosesan data seperti pembersihan data, normalisasi teks, segmentasi karakter, penambahan karakter pengisi (padding), serta pembagian dataset menjadi data pelatihan, validasi, dan pengujian. Model transliterasi dibangun menggunakan arsitektur encoder-decoder berbasis LSTM. Mekanisme Attention diintegrasikan ke dalam model agar dapat mempelajari hubungan ketergantungan antara urutan karakter masukan dan keluaran secara lebih efektif.
Evaluasi model dilakukan menggunakan metrik Character Error Rate (CER) dan Word Error Rate (WER), serta dibandingkan dengan model Seq2Seq LSTM tanpa mekanisme Attention. Hasil pengujian menunjukkan bahwa model Seq2Seq LSTM dengan mekanisme Attention memberikan performa yang lebih baik dengan nilai CER sebesar 2,23%, WER sebesar 9,52%, dan akurasi exact-match sebesar 52,94%. Hasil tersebut menunjukkan bahwa mekanisme Attention mampu meningkatkan kualitas transliterasi aksara Jawa. Penelitian ini diharapkan dapat berkontribusi dalam mendukung digitalisasi manuskrip aksara Jawa sehingga hasil transliterasi lebih mudah dipahami dan dimanfaatkan oleh masyarakat maupun peneliti di bidang kebudayaan dan humaniora.
Kata kunci: Aksara Jawa, Bahdanau Attention, LSTM, Neural Mechine Translation, Sequence-to-Sequence, Transliterasi.
====================================================================================================================================
The Javanese script is a cultural heritage of the Indonesian archipelago widely found in various ancient manuscripts. Although previous research has successfully detected Javanese characters from manuscript images, the outputs remain in sequences of character classes that are difficult for laypersons to read. Transliterating Javanese script into the Latin alphabet remains a challenge due to its abugida nature and complex inter-character relationships. To address this issue, this study developed a Javanese script transliteration system using a Sequence-to-Sequence (Seq2Seq) model with Long Short-Term Memory (LSTM) and the Bahdanau Attention mechanism. The dataset utilized consists of pairs of Javanese character sequences and their corresponding Latin transliterations from prior research.
The methodology begins with data preprocessing, including data cleaning, text normalization, character segmentation, padding, and splitting the dataset into training, validation, and test sets. The transliteration model is constructed using an LSTM-based encoder-decoder architecture. An attention mechanism is integrated to enable the model to learn the mapping and relationships between input and output character sequences more effectively.
The model was evaluated using Character Error Rate (CER) and Word Error Rate (WER) metrics and compared with a Seq2Seq LSTM model without an Attention mechanism. The experimental results show that the Seq2Seq LSTM model with the Attention mechanism achieved better performance, obtaining a CER of 2.23%, a WER of 9.52%, and an exact-match accuracy of 52.94%. These findings indicate that the Attention mechanism improves the quality of Javanese script transliteration. This research is expected to contribute to the digitalization of Javanese manuscripts by producing transliteration results that are easier to understand and utilize for both the general public and researchers in the fields of culture and humanities.
Keywords: Bahdanau Attention, Javanese Script, LSTM, Neural Machine Translation, Sequence-to-Sequence, Transliteration.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Aksara Jawa, Bahdanau Attention, LSTM, Neural Mechine Translation, Sequence-to-Sequence, Transliterasi, Bahdanau Attention, Javanese Script, LSTM, Neural Machine Translation, Sequence-to-Sequence, Transliteration
Subjects:	Q Science T Technology > T Technology (General)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Ike Norfaize
Date Deposited:	25 Jul 2026 09:02
Last Modified:	25 Jul 2026 09:02
URI:	http://repository.its.ac.id/id/eprint/138277

Actions (login required)

View Item