ASPECT BASED SENTIMENT ANALYSIS UNTUK MENGIDENTIFIKASI KELUHAN PENAGIHAN PINJAMAN ONLINE DI MEDIA SOSIAL X MENGGUNAKAN ALGORITMA BAYESIAN NETWORK DAN C4.5

FRASETIA ADI KUSUMA, . (2026) ASPECT BASED SENTIMENT ANALYSIS UNTUK MENGIDENTIFIKASI KELUHAN PENAGIHAN PINJAMAN ONLINE DI MEDIA SOSIAL X MENGGUNAKAN ALGORITMA BAYESIAN NETWORK DAN C4.5. Sarjana thesis, UNIVERSITAS NEGERI JAKARTA.

	Text Cover.pdf Download (473kB)
	Text BAB I.pdf Download (364kB)
	Text BAB II.pdf Restricted to Registered users only Download (766kB) | Request a copy
	Text BAB III.pdf Restricted to Registered users only Download (481kB) | Request a copy
	Text BAB IV.pdf Restricted to Registered users only Download (1MB) | Request a copy
	Text BAB V.pdf Restricted to Registered users only Download (295kB) | Request a copy
	Text Daftar Pustaka.pdf Download (277kB)
	Text Lampiran.pdf Restricted to Registered users only Download (1MB) | Request a copy

Abstract

Penelitian ini mengimplementasikan Aspect-Based Sentiment Analysis (ABSA) keluhan penagihan pinjaman online di media sosial X dengan membandingkan algoritma Bayesian Network dan C4.5. Data diproses mengacu pada tahapan Knowledge Discovery in Databases (KDD) hingga tahap representasi vektor numerik menggunakan FastText. Teknik Synthetic Minority Over-sampling Technique (SMOTE) diterapkan pada keseluruhan data untuk menangani ketidakseimbangan kelas, kemudian dievaluasi menggunakan skema Stratified 5-Fold Cross Validation. Hasil dari delapan skenario pengujian menunjukkan bahwa intervensi SMOTE justru menurunkan kinerja Bayesian Network akibat distorsi probabilitas. Hal ini ditandai dengan turunnya akurasi aspek dari 75,50% menjadi 75,20%, serta meningkatnya angka false positive pada sentimen, meskipun akurasi bergeser dari 73,10% ke 76,50%. Sebaliknya, performa algoritma C4.5 meningkat secara masif setelah penggunaan SMOTE, dengan lonjakan akurasi pada kelas aspek 76,30% menjadi 84,00% dan target sentimen dari 74,80% menjadi 81,90%. Kesimpulannya, algoritma C4.5 terbukti lebih adaptif dalam memotong batas keputusan pada ruang vektor dense FastText. Integrasi FastText, SMOTE, Stratified 5-Fold CV, dan C4.5 merupakan arsitektur pemodelan terbaik untuk mengklasifikasikan teks keluhan praktik penagihan pada penelitian ini. ***** This study implements Aspect-Based Sentiment Analysis (ABSA) on online loan collection complaints on social media X by comparing Bayesian Network and C4.5 algorithms. Data processing follows the Knowledge Discovery in Databases (KDD) stages up to the numerical vector representation stage using FastText. The Synthetic Minority Over-sampling Technique (SMOTE) was applied to the entire dataset to handle class imbalance, followed by an evaluation using a Stratified 5-Fold Cross Validation scheme. The results from eight testing scenarios indicate that SMOTE intervention actually degraded the performance of the Bayesian Network due to probability distortion. This was marked by a decline in aspect target accuracy from 75.50% to 75.20%, along with an increase in false positive rates for the sentiment target, despite the overall accuracy shifting from 73.10% to 76.50%. Conversely, the performance of the C4.5 algorithm improved massively following the application of SMOTE, with a surge in accuracy for the aspect target from 76.30% to 84.00% and the sentiment target from 74.80% to 81.90%. In conclusion, the C4.5 algorithm proved to be more adaptive in establishing decision boundaries within the FastText dense vector space. The integration of FastText, SMOTE, Stratified 5-Fold CV, and C4.5 constitutes the optimal modeling architecture for classifying debt collection complaint texts in this study.

Item Type:	Thesis (Sarjana)
Additional Information:	1). Dr. Widodo, S.Kom., M.Kom. ; 2). Neng Ayu Herawati, S.Pd., M.T.
Subjects:	Sains > Matematika > Ilmu Komputer Teknologi dan Ilmu Terapan > Teknik Komputer
Divisions:	FT > S1 Pendidikan Teknik Informatika Komputer
Depositing User:	Frasetia Adi Kusuma .
Date Deposited:	12 Aug 2026 07:52
Last Modified:	12 Aug 2026 07:52
URI:	http://repository.unj.ac.id/id/eprint/67461

Actions (login required)

View Item