underserved lintas klaster dengan skor peluang tertinggi mencapai 17,471 pada segmen premium dan 12,130 pada segmen ekonomis, sementara aspek Pengalaman Aplikasi (desain fisik aplikator) menjadi fokus kritis underserved khusus pada Klaster III dengan skor kepuasan terendah sebesar 1,000. The cosmetics industry in Indonesia is experiencing rapid growth, triggering intense competition, particularly within the lip cream product category. In this dynamic environment, companies are compelled to deeply understand consumer needs to minimize the risk of new product failure. However, consumer reviews on public platforms such as Female Daily are often massive in volume and exist in unstructured text formats containing slang and abbreviations, making manual analysis highly difficult. This study aims to automatically and accurately extract the "Voice of Customer" (VoC) to identify the most crucial product attributes for lip cream development and marketing strategies.The methodology of this research employs a big data analytics approach by exclusively extracting 47,737 consumer reviews from 216 Stock Keeping Units (SKUs) on the Female Daily Network platform within the 2022 to 2025 period. The data were cleaned through preprocessing stages and subjected to a data adequacy test utilizing a formula with a 5% precision level and a 95% confidence level. The data were then grouped into four cluster quadrants based on the variables of standardized average price per milliliter and product rating as the industry baseline. The VoC extraction process was conducted using the probabilistic Latent Dirichlet Allocation (LDA) algorithm, which was subsequently integrated with the Transformer-based Deep Learning model, BERT, at the sentence level to evaluate customer satisfaction rates.The computational experiment results demonstrate optimal LDA model performance, with an average Coherence Score within an excellent range (0.42 to 0.58) and a low Jaccard Similarity overlap value (< 0>"> underserved lintas klaster dengan skor peluang tertinggi mencapai 17,471 pada segmen premium dan 12,130 pada segmen ekonomis, sementara aspek Pengalaman Aplikasi (desain fisik aplikator) menjadi fokus kritis underserved khusus pada Klaster III dengan skor kepuasan terendah sebesar 1,000. The cosmetics industry in Indonesia is experiencing rapid growth, triggering intense competition, particularly within the lip cream product category. In this dynamic environment, companies are compelled to deeply understand consumer needs to minimize the risk of new product failure. However, consumer reviews on public platforms such as Female Daily are often massive in volume and exist in unstructured text formats containing slang and abbreviations, making manual analysis highly difficult. This study aims to automatically and accurately extract the "Voice of Customer" (VoC) to identify the most crucial product attributes for lip cream development and marketing strategies.The methodology of this research employs a big data analytics approach by exclusively extracting 47,737 consumer reviews from 216 Stock Keeping Units (SKUs) on the Female Daily Network platform within the 2022 to 2025 period. The data were cleaned through preprocessing stages and subjected to a data adequacy test utilizing a formula with a 5% precision level and a 95% confidence level. The data were then grouped into four cluster quadrants based on the variables of standardized average price per milliliter and product rating as the industry baseline. The VoC extraction process was conducted using the probabilistic Latent Dirichlet Allocation (LDA) algorithm, which was subsequently integrated with the Transformer-based Deep Learning model, BERT, at the sentence level to evaluate customer satisfaction rates.The computational experiment results demonstrate optimal LDA model performance, with an average Coherence Score within an excellent range (0.42 to 0.58) and a low Jaccard Similarity overlap value (< 0>">
Laporkan Masalah

INTEGRASI PEMODELAN TOPIK LDA, ANALISIS SENTIMEN BERT, DAN ALGORITMA PELUANG DALAM EKSTRAKSI SUARA PELANGGAN STUDI KASUS: PENGEMBANGAN PRODUK LIP CREAM

Vannencia Magdalena, Prof. Ir. Subagyo, Ph.D., IPU., ASEAN Eng.

2026 | Tesis | S2 Teknik Industri

Industri kosmetik di Indonesia mengalami pertumbuhan pesat yang memicu persaingan ketat, khususnya pada kategori produk lip cream. Dalam lingkungan yang dinamis ini, perusahaan dituntut untuk memahami kebutuhan konsumen secara mendalam guna meminimalkan risiko kegagalan produk baru. Namun, ulasan konsumen pada platform publik seperti Female Daily sering kali hadir dalam jumlah masif dan format teks tidak terstruktur yang mengandung bahasa gaul (slang) serta singkatan, sehingga sulit dianalisis secara manual. Penelitian ini bertujuan untuk mengekstraksi "Suara Pelanggan" (Voice of Customer) secara otomatis dan akurat untuk mengidentifikasi atribut produk yang paling krusial bagi pengembangan dan strategi pemasaran lip cream.

Metode penelitian ini menggunakan pendekatan big data analytics dengan mengekstrak 47.737 ulasan konsumen dari 216 SKU pada platform Female Daily Network secara eksklusif dalam rentang tahun 2022 hingga 2025. Data dibersihkan melalui tahap preprocessing dan diuji kecukupan datanya menggunakan rumus tingkat ketelitian 5?n kepercayaan 95%. Data kemudian dikelompokkan ke dalam empat kuadran klaster berdasarkan variabel rata-rata harga terstandardisasi per mililiter dan rating produk sebagai industry baseline. Proses ekstraksi VoC dilakukan menggunakan algoritma probabilistik Latent Dirichlet Allocation (LDA), yang kemudian diintegrasikan dengan model Deep Learning berbasis Transformer BERT pada level kalimat untuk menilai tingkat kepuasan pelanggan.

Hasil eksperimen komputasi menunjukkan performansi model LDA yang optimal dengan rata-rata Coherence Score pada rentang sangat baik (0,42 hingga 0,58) dan nilai tumpang tindih Jaccard Similarity yang rendah (< 0>underserved lintas klaster dengan skor peluang tertinggi mencapai 17,471 pada segmen premium dan 12,130 pada segmen ekonomis, sementara aspek Pengalaman Aplikasi (desain fisik aplikator) menjadi fokus kritis underserved khusus pada Klaster III dengan skor kepuasan terendah sebesar 1,000.

The cosmetics industry in Indonesia is experiencing rapid growth, triggering intense competition, particularly within the lip cream product category. In this dynamic environment, companies are compelled to deeply understand consumer needs to minimize the risk of new product failure. However, consumer reviews on public platforms such as Female Daily are often massive in volume and exist in unstructured text formats containing slang and abbreviations, making manual analysis highly difficult. This study aims to automatically and accurately extract the "Voice of Customer" (VoC) to identify the most crucial product attributes for lip cream development and marketing strategies.

The methodology of this research employs a big data analytics approach by exclusively extracting 47,737 consumer reviews from 216 Stock Keeping Units (SKUs) on the Female Daily Network platform within the 2022 to 2025 period. The data were cleaned through preprocessing stages and subjected to a data adequacy test utilizing a formula with a 5% precision level and a 95% confidence level. The data were then grouped into four cluster quadrants based on the variables of standardized average price per milliliter and product rating as the industry baseline. The VoC extraction process was conducted using the probabilistic Latent Dirichlet Allocation (LDA) algorithm, which was subsequently integrated with the Transformer-based Deep Learning model, BERT, at the sentence level to evaluate customer satisfaction rates.

The computational experiment results demonstrate optimal LDA model performance, with an average Coherence Score within an excellent range (0.42 to 0.58) and a low Jaccard Similarity overlap value (< 0>


Kata Kunci : Pemodelan Topik, LDA, Analisis Sentimen, BERT, Opportunity Algorithm, Lip Cream

  1. S2-2026-541267-abstract.pdf  
  2. S2-2026-541267-bibliography.pdf  
  3. S2-2026-541267-tableofcontent.pdf  
  4. S2-2026-541267-title.pdf