ANALISIS PENGENDALIAN KUALITAS PRODUK BAKPIA KERING DENGAN METODE STATISTICAL QUALITY CONTROL PADA UMKM BAKPIA PATHOK 52 KOTA YOGYAKARTA

HIDAYAH, NURUL (2025) ANALISIS PENGENDALIAN KUALITAS PRODUK BAKPIA KERING DENGAN METODE STATISTICAL QUALITY CONTROL PADA UMKM BAKPIA PATHOK 52 KOTA YOGYAKARTA. Skripsi thesis, UPN Veteran Yogyakarta.

	Text 1_Skripsi_Full_123210114_Rizal_Hanifa_Pratama.pdf Restricted to Repository staff only Download (7MB)
	Text 2_Cover_123210114_Rizal_Hanifa_Pratama.pdf Download (431kB)
	Text 3_Abstrak_123210114_Rizal_Hanifa_Pratama.pdf Download (246kB)
	Text 4_Halaman_Pengesahan_123210114_Rizal_Hanifa_Pratama.pdf Download (943kB)
	Text 5_Daftar_Isi_123210114_Rizal_Hanifa_Pratama.pdf Download (358kB)
	Text 6_Daftar_Pustaka_123210114_Rizal_Hanifa_Pratama.pdf Download (196kB)

Abstract

ABSTRAK
Permasalahan utama dalam penelitian ini adalah bagaimana meningkatkan performa
klasifikasi sentimen terhadap ulasan pelanggan restoran, khususnya dalam konteks bahasa
Indonesia. Ulasan pelanggan mengandung ekspresi opini yang kompleks dan sering kali
tidak eksplisit, sehingga memerlukan pendekatan pemrosesan bahasa alami yang mampu
menangkap nuansa sentimen secara akurat. Model IndoBERT sebagai representasi
transformer-based language model berbahasa Indonesia telah menunjukkan performa yang
baik dalam berbagai tugas pemahaman bahasa, namun belum dioptimalkan secara spesifik
untuk domain ulasan restoran. Oleh karena itu, penelitian ini berfokus pada pengaruh
penggunaan fitur N-Gram dan optimasi hyperparameter dalam proses fine-tuning model
IndoBERT untuk meningkatkan akurasi klasifikasi sentimen tiga kelas pada data ulasan
restoran.
Metode penelitian ini melibatkan beberapa tahapan utama, dimulai dari
pengumpulan data ulasan pelanggan dari Restoran Kampung Kecil Semarang. Data
kemudian dilabeli secara otomatis menggunakan lexicon INSET untuk klasifikasi sentimen
positif, netral, dan negatif. Dua pendekatan pelabelan diterapkan, yaitu dengan integrasi N
Gram dan tanpa N-Gram. Model IndoBERT base-p1 digunakan sebagai fondasi, yang
kemudian melalui proses fine-tuning dengan skema pencarian hyperparameter
menggunakan metode Grid Search. Ruang pencarian dirancang berdasarkan rekomendasi
dari Devlin et al., dengan mempertimbangkan kombinasi nilai batch size, learning rate, dan
jumlah epoch. Evaluasi performa model dilakukan menggunakan metrik akurasi dan F1
Score.
Hasil penelitian menunjukkan bahwa penggunaan N-Gram memberikan dampak
positif yang konsisten terhadap peningkatan performa model, baik pada tahap pre-trained
maupun setelah fine-tuning. Model dengan N-Gram yang telah di-fine-tune mencapai
akurasi sebesar 0,924429 dan F1-Score sebesar 0,924816, sedangkan model tanpa N-Gram
hanya mencapai akurasi 0,846834 dan F1-Score 0,846685. Peningkatan performa ini
menunjukkan bahwa N-Gram efektif dalam menangkap konteks frasa dan polaritas
gabungan yang penting untuk tugas klasifikasi sentimen. Selain itu, proses fine-tuning
dengan optimasi hyperparameter terbukti meningkatkan kinerja model secara signifikan
dibandingkan model pre-trained. Penelitian ini memberikan kontribusi terhadap
pengembangan sistem analisis sentimen yang lebih adaptif terhadap domain spesifik, serta
mempertegas pentingnya pemanfaatan fitur linguistik tambahan dalam meningkatkan
akurasi model berbasis transformer.
Kata Kunci: klasifikasi sentimen, IndoBERT, N-Gram, fine-tuning, optimasi
hyperparameter
v

ABSTRACT
The main problem addressed in this study is how to improve sentiment classification
performance on customer reviews in the context of the Indonesian language. Customer
reviews often contain complex and implicit expressions of opinion, requiring a natural
language processing approach that can accurately capture sentiment nuances. IndoBERT,
as a transformer-based language model for Indonesian, has demonstrated promising
performance across various language understanding tasks, yet it has not been specifically
optimized for the restaurant review domain. Therefore, this research focuses on investigating
the impact of N-Gram feature integration and hyperparameter optimization in the fine
tuning process of the IndoBERT model to enhance the accuracy of three-class sentiment
classification on restaurant review data.
The research methodology involves several main stages, beginning with the
collection of customer reviews from Restoran Kampung Kecil Semarang. The data were
automatically labeled using the INSET lexicon for positive, neutral, and negative sentiment
categories. Two labeling approaches were applied: one with N-Gram integration and one
without. The IndoBERT base-p1 model was used as the foundation, which was then fine
tuned through Grid Search hyperparameter optimization. The search space was constructed
based on recommendations from Devlin et al., considering combinations of batch size,
learning rate, and the number of epochs. Model performance was evaluated using accuracy
and F1-Score metrics.
The results show that the use of N-Gram has a consistently positive impact on model
performance, both in the pre-trained stage and after fine-tuning. The fine-tuned model with
N-Gram achieved an accuracy of 0.924429 and an F1-Score of 0.924816, while the model
without N-Gram achieved only 0.846834 and 0.846685, respectively. This performance
improvement indicates that N-Gram is effective in capturing phrase-level context and
combined sentiment polarity, which are crucial for sentiment classification tasks.
Additionally, the fine-tuning process with hyperparameter optimization significantly
improves the model’s performance compared to the pre-trained version. This study
contributes to the development of sentiment analysis systems that are more adaptive to
specific domains and underscores the importance of incorporating additional linguistic
features to enhance the accuracy of transformer-based models.
Keywords: sentiment classification, IndoBERT, N-Gram, fine-tuning, hyperparameter
optimization
vi

Item Type:	Tugas Akhir (Skripsi)
Uncontrolled Keywords:	sentiment classification, IndoBERT, N-Gram, fine-tuning, hyperparameter optimization
Subjek:	S Agriculture > S Agriculture (General)
Divisions:	Fakultas Pertanian > (S1) Agribisnis
Depositing User:	Eko Yuli
Date Deposited:	27 Aug 2025 01:54
Last Modified:	27 Aug 2025 01:54
URI:	http://eprints.upnyk.ac.id/id/eprint/43538

Actions (login required)

View Item