Perbandingan OPTICS dan HDBSCAN Pada Algoritma BERTopic Untuk Topic Modeling Berbasis Analisis Sentimen Pada Pengguna X Terhadap Kinerja Kepolisian

Razan, Akbar (2026) Perbandingan OPTICS dan HDBSCAN Pada Algoritma BERTopic Untuk Topic Modeling Berbasis Analisis Sentimen Pada Pengguna X Terhadap Kinerja Kepolisian. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221150-Undergradute_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (10MB) | Request a copy

Abstract

Transisi kepemimpinan nasional dari pemerintahan Joko Widodo menuju Prabowo Subianto mendorong pergeseran fokus strategis Polri yang memicu dinamika evaluasi publik di media sosial X. Penelitian ini mengidentifikasi karakteristik data, mengklasifikasikan polaritas sentimen, serta memetakan topik perdebatan mengenai kinerja kepolisian pada periode Juli 2023 hingga Desember 2025 menggunakan model IndoBERTweet dan kerangka BERTopic berbasis 4.618 dokumen yang dikumpulkan melalui pencarian kata kunci "kinerja polisi". Hasil analisis karakteristik data menunjukkan volume kritik masyarakat mengalami lonjakan tajam pascapelantikan, sementara respons positif stagnan. Evaluasi klasifikasi sentimen menunjukkan model IndoBERTweet efektif dengan akurasi 93,40% serta F1-Score makro 91,03% dalam memisahkan 3.480 dokumen negatif dan 1.138 dokumen positif. Dalam pemodelan topik, algoritma HDBSCAN lebih representatif dibanding OPTICS dengan proporsi outlier yang jauh lebih rendah, sebesar 28,73% berbanding 85,94% pada data positif serta 62,36% berbanding 85,63% pada data negatif, dengan koherensi sebesar 0,54 dan topic diversity sebesar 0,97 untuk dokumen bersentimen positif, dan koherensi sebesar 0,48 dan topic diversity sebesar 0,90 untuk dokumen bersentimen negatif. Ekstraksi HDBSCAN memetakan tujuh topik positif yang terpusat pada apresiasi kinerja operasional dan pelayanan publik, serta delapan topik negatif yang didominasi tuntutan reformasi institusional dan erosi kepercayaan masyarakat, dengan pola temporal yang bergeser dari perbandingan institusi luar negeri di awal periode menuju kritik anggaran dan isu-isu spesifik seperti penanganan kasus berbasis viralitas dan skeptisisme survei kepuasan pada akhir periode pengamatan. Penelitian ini menunjukkan bahwa pendekatan IndoBERTweet dan BERTopic mampu mengungkap dinamika opini publik terhadap institusi kepolisian secara sistematis dan dapat menjadi rujukan bagi pemangku kebijakan dalam memantau persepsi masyarakat berbasis data media sosial.
=====================================================================================================================================
The national leadership transition from Joko Widodo to Prabowo Subianto shifted Polri's strategic focus and triggered public evaluation dynamics on social media X. This study identifies data characteristics, classifies sentiment polarity, and maps debate topics on police performance from July 2023 to December 2025 using IndoBERTweet and BERTopic based on 4,618 documents collected through the keyword "kinerja polisi". Analysis of data characteristics reveals that the volume of public criticism surged sharply after the inauguration, while positive responses remained stagnant. Sentiment classification shows IndoBERTweet to be effective, achieving 93.40% accuracy and 91.03% macro F1-Score in separating 3,480 negative and 1,138 positive documents. In topic modeling, HDBSCAN outperforms OPTICS in representativeness with substantially lower outlier proportions of 28.73% versus 85.94% on positive data and 62.36% versus 85.63% on negative data, with coherence of 0.54 and topic diversity of 0.97 for positive sentiment documents, and coherence of 0.48 and topic diversity of 0.90 for negative sentiment documents. HDBSCAN extracted seven positive topics centered on appreciation of operational performance and public services, and eight negative topics dominated by demands for institutional reform and erosion of public trust, with a temporal pattern shifting from comparisons with foreign institutions in the early period toward budget criticism and specific issues such as virality-driven case handling and skepticism toward satisfaction surveys toward the end of the observation period. This study demonstrates that the IndoBERTweet and BERTopic approach is capable of systematically uncovering public opinion dynamics toward law enforcement institutions and can serve as a reference for policymakers in monitoring public perception from social media data.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Analisis Sentimen, BERTopic, IndoBERTweet, Kinerja Kepolisian, OPTICS, BERTopic, IndoBERTweet, OPTICS, Police Performance, Sentiment Analysis
Subjects:	H Social Sciences > HM Sociology > HM742 Online social networks. Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA278.55 Cluster analysis Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science)
Divisions:	Faculty of Mathematics and Science > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Akbar Razan
Date Deposited:	30 Jul 2026 06:31
Last Modified:	30 Jul 2026 06:31
URI:	http://repository.its.ac.id/id/eprint/140180

Actions (login required)

View Item