ANALISIS SENTIMEN ULASAN PRODUK MENGGUNAKAN LARGE LANGUAGE MODELS: STUDI KASUS PADA SHOPEE
Abstract
Perkembangan pesat e-commerce telah meningkatkan jumlah ulasan produk yang tersedia di platform seperti Shopee. Ulasan ini dapat menjadi sumber informasi berharga bagi penjual dalam memahami persepsi konsumen dan meningkatkan strategi pemasaran. Namun, besarnya volume dan kompleksitas bahasa dalam ulasan membuat analisis manual menjadi tidak efisien. Penelitian ini bertujuan untuk menganalisis sentimen ulasan produk di Shopee menggunakan Large Language Models (LLMs), khususnya model Gemini 1.5-Pro yang telah di-fine-tune agar lebih sesuai dengan bahasa pengguna Shopee. Metode yang digunakan mencakup pengumpulan data melalui web scraping, preprocessing data, fine-tuning model, serta evaluasi performa model berdasarkan metrik akurasi, presisi, recall, dan F1-score. Hasil penelitian menunjukkan bahwa model yang dikembangkan mampu mengklasifikasikan sentimen ulasan ke dalam kategori positif, negatif, dan netral dengan akurasi berkisar antara 66,67% hingga 85,71%.
Published
How to Cite
Issue
Section
License
Copyright (c) 2025 Ribka Amelia Yunita Keliat, Evta Indra, Yonata Laia

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Authors who publish their manuscripts through the Journal of Information Systems and Computer Science agree to the following:
- Copyright to the manuscripts of scientific papers in this Journal is held by the author.
- The author surrenders the rights when first publishing the manuscript of his scientific work and simultaneously the author grants permission / license by referring to the Creative Commons Attribution-ShareAlike 4.0 International License to other parties to distribute his scientific work while still giving credit to the author and the Journal of Information Systems and Computer Science as the first publication medium for the work.
- Matters relating to the non-exclusivity of the distribution of the Journal that publishes the author's scientific work can be agreed separately (for example: requests to place the work in the library of an institution or publish it as a book) with the author as one of the parties to the agreement and with credit to sJournal of Information Systems and Computer Science as the first publication medium for the work in question.
- Authors can and are expected to publish their work online (e.g. in a Repository or on their Organization's/Institution's website) before and during the manuscript submission process, as such efforts can increase citation exchange earlier and with a wider scope.