<![CDATA[Abstract: The replacement of Indonesia's Minister of Finance from Sri Mulyani Indrawati to Purbaya Yudhi Sadewa on September 8, 2025, triggered significant public discourse regarding institutions under the Ministry of Finance, particularly the Directorate General of Customs and Excise (DJBC). This research aims to identify and classify dominant topics discussed by the public on social media regarding Customs during the period from September 2025 to April 2026 using the Latent Dirichlet Allocation (LDA) algorithm. A total of 937 public comments were collected through web scraping on YouTube from 15 relevant news videos, followed by preprocessing including case folding, tokenization, stopword removal using the Sastrawi library, and stemming. The optimal number of topics was determined by evaluating Coherence Score (C_v) and Log Perplexity across the range K = 2 to K = 15. The results indicate that the highest coherence value was obtained at K = 11 with a score of 0.3436, which was selected as the final model. The eleven topics generated reflect a spectrum of public opinion covering appreciation for Minister Purbaya, demands for cleansing corrupt officials, illegal smuggling issues, and demands for institutional reform. Topic distribution shows that 52.3% of documents were critical, 29.9% supportive, and 18.1% reform-demanding. These findings can serve as a basis for evaluating DJBC's public communication policies under Minister Purbaya's leadership. Keyword: latent dirichlet allocation; topic modeling; customs; social media; coherence score. Abstrak: Pergantian Menteri Keuangan Republik Indonesia dari Sri Mulyani Indrawati kepada Purbaya Yudhi Sadewa pada 8 September 2025 memicu dinamika pemberitaan signifikan terhadap institusi-institusi di bawah Kementerian Keuangan, khususnya Direktorat Jenderal Bea dan Cukai (DJBC). Penelitian ini bertujuan untuk mengidentifikasi dan mengklasifikasikan topik-topik dominan yang diperbincangkan publik di media sosial terkait Bea Cukai pada periode September 2025 hingga April 2026 menggunakan algoritma Latent Dirichlet Allocation (LDA). Data sejumlah 937 komentar publik dikumpulkan melalui teknik web scraping pada platform YouTube dari 15 video pemberitaan relevan, kemudian diolah melalui tahapan preprocessing berupa case folding, tokenisasi, stopword removal menggunakan library Sastrawi, dan stemming. Penentuan jumlah topik optimal dilakukan dengan evaluasi Coherence Score (C_v) dan Log Perplexity pada rentang K = 2 hingga K = 15. Hasil penelitian menunjukkan bahwa nilai coherence tertinggi diperoleh pada K = 11 dengan skor 0,3436, yang selanjutnya dijadikan sebagai model final. Sebelas topik yang dihasilkan mencerminkan spektrum opini publik yang terdiri dari topik apresiasi terhadap Menkeu Purbaya, tuntutan pembersihan oknum korup, isu penyelundupan barang ilegal, serta desakan reformasi kelembagaan. Distribusi topik menunjukkan bahwa 52,3% dokumen bersifat kritis, 29,9% bersifat dukungan, dan 18,1% berupa desakan reformasi. Temuan ini dapat dimanfaatkan sebagai dasar evaluasi kebijakan komunikasi publik DJBC di era kepemimpinan Menkeu Purbaya. Kata kunci: bea cukai; coherence score; latent dirichlet allocation; media sosial; pemodelan topik]]>
