Analisis Sentimen Komentar Youtube terhadap Kondisi Bursa Saham Indonesia akibat Isu Pengunduran Serempak Dewan BEI Menggunakan IndoBERT

Abstract

Social media platforms such as YouTube have long served as a primary discussion space for retail investor communities in Indonesia. This study aims to analyze public sentiment in order to understand perception trends and the digital psychology of capital market participants regarding the issue of the simultaneous resignation of the Indonesia Stock Exchange (IDX) board members. The research applies the IndoBERT (Bidirectional Encoder Representations from Transformers for the Indonesian language) deep learning architecture through a fine-tuning process on a dataset of YouTube comments. The textual corpus was cleaned from noise, normalized from stock market slang vocabulary, tokenized, and automatically classified into three sentiment polarities: positive, neutral, and negative. The analysis stage was further continued with dominant keyword extraction using Word Cloud visualization and word frequency trend mapping to identify psychological variables driving market opinions. The model successfully classified the semantic complexity of informal language objectively. Visualization results indicate that communication dynamics were overwhelmingly dominated by negative sentiment (57.5%), reflecting widespread public concern and declining confidence in capital market stability due to the structural crisis. This study demonstrates the effectiveness of local transformer models as instruments for extracting digital market psychology to support real-time automated investment decision-making.Platform media sosial seperti YouTube sudah lama menjadi wadah utama diskusi bagi komunitas investor ritel di Indonesia. Penelitian ini bertujuan menganalisis sentimen publik untuk memahami tren persepsi dan psikologi digital para pelaku pasar modal terhadap isu pengunduran diri serempak dewan Bursa Efek Indonesia (BEI). Penelitian ini menerapkan arsitektur Deep Learning IndoBERT (Bidirectional Encoder Representations from Transformers untuk bahasa Indonesia) melalui proses fine-tuning pada dataset komentar YouTube. Korpus data tekstual dibersihkan dari derau (noise), dinormalisasi dari kosakata slang pasar modal, ditokenisasi, dan diklasifikasikan secara otomatis ke dalam tiga polaritas sentimen: positif, netral, dan negatif. Tahap analisis dilanjutkan dengan ekstraksi kata kunci dominan berbasis Word Cloud serta pemetaan tren frekuensi kata untuk mengidentifikasi variabel psikologis yang menggerakkan opini pasar. Model berhasil mengklasifikasikan kompleksitas semantik bahasa informal secara objektif. Hasil visualisasi menunjukkan bahwa dinamika komunikasi didominasi secara mutlak oleh sentimen negatif (57,5%), mencerminkan adanya keresahan massal dan penurunan tingkat kepercayaan terhadap stabilitas pasar modal akibat krisis struktural tersebut. Penelitian ini membuktikan efektivitas model transformer lokal sebagai instrumen ekstraksi psikologi pasar digital guna mendukung otomatisasi pengambilan keputusan investasi secara real-time.