MA'ARUF, AMIL (2026) ANALISIS SENTIMEN MULTIMODAL PADA PLATFORM X DAN BERITA MAKROEKONOMI MENGGUNAKAN MODEL FINROBERTA UNTUK MENGETAHUI ARAH HARGA BITCOIN. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
Cover.pdf Download (423kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
BAB 1.pdf Restricted to Registered users only Download (249kB) |
|
|
Text (BAB II)
BAB 2.pdf Restricted to Registered users only Download (592kB) |
|
|
Text (BAB III)
BAB 3.pdf Restricted to Registered users only Download (334kB) |
|
|
Text (BAB IV)
BAB 4.pdf Restricted to Registered users only Download (5MB) |
|
|
Text (BAB V)
BAB 5.pdf Restricted to Registered users only Download (213kB) |
|
|
Text (DAFTAR PUSTAKA)
Daftar Pustaka.pdf Restricted to Registered users only Download (255kB) |
|
|
Text (LAMPIRAN)
Lampiran.pdf Restricted to Registered users only Download (233kB) |
Abstract
Bitcoin's price volatility is significantly influenced by public sentiment on social media and macroeconomic news releases. Previous studies generally analyzed these two data sources separately (single-source), failing to capture a comprehensive sentiment signal to confirm Bitcoin's price direction. This study aims to examine the ability of the FinRoBERTa model to convert multimodal textual data (Platform X tweets and macroeconomic news) into quantitative sentiment indicators, evaluate the effect of combining two data sources on classification accuracy and Bitcoin price direction, and analyze the effect of LoRA on computational efficiency and model accuracy. This study employs a quantitative method using secondary data consisting of 126,375 Twitter tweets, 36,280 news articles, and historical Bitcoin price data (OHLCV) obtained from the Binance API. FinRoBERTa was optimized using LoRA (rank=8, alpha=32) for three-class sentiment classification, which was subsequently mapped by XGBoost to predict Bitcoin's price direction (Up, Down, Sideways). The results show that FinRoBERTa achieved an accuracy of 93.32% on the combined dataset, outperforming BERT (92.40%) and FinBERT (92.90%). Combining the two data sources did not improve classification accuracy compared to using Twitter data alone (95.97%), yet the combined sentiment signal remained beneficial as a predictive signal for price direction through XGBoost, achieving an accuracy of 61.11%, well above the random baseline of 33%. The application of LoRA proved efficient, training only 0.51% of FinRoBERTa's total parameters without compromising model accuracy. This study concludes that multimodal sentiment integration combined with LoRA optimization can produce accurate and computationally efficient sentiment signals to support Bitcoin price direction prediction. Keywords: FinRoBERTa, LoRA, multimodal sentiment analysis, XGBoost, Bitcoin price direction. Volatilitas harga Bitcoin sangat dipengaruhi oleh sentimen publik di media sosial dan rilis berita makroekonomi global. Penelitian terdahulu umumnya menganalisis kedua sumber data tersebut secara terpisah (single-source), sehingga belum menangkap sinyal sentimen yang komprehensif untuk mengonfirmasi arah pergerakan harga Bitcoin. Penelitian ini bertujuan menguji kemampuan model FinRoBERTa dalam mengonversi data teks multimodal (cuitan Platform X dan berita makroekonomi) menjadi indikator sentimen kuantitatif, mengevaluasi pengaruh penggabungan dua sumber data terhadap akurasi klasifikasi dan arah harga Bitcoin, serta menganalisis pengaruh LoRA terhadap efisiensi komputasi dan akurasi model. Penelitian menggunakan metode kuantitatif dengan data sekunder berupa 126.375 cuitan Twitter, 36.280 artikel berita, dan data historis harga Bitcoin (OHLCV) dari API Binance. FinRoBERTa dioptimasi dengan LoRA (rank=8, alpha=32) untuk klasifikasi sentimen tiga kelas, yang selanjutnya dipetakan oleh XGBoost untuk memprediksi arah harga Bitcoin (Up, Down, Sideways). Hasil penelitian menunjukkan FinRoBERTa mencapai akurasi 93,32% pada data gabungan, lebih tinggi dibanding BERT (92,40%) dan FinBERT (92,90%). Penggabungan dua sumber data tidak meningkatkan akurasi klasifikasi dibanding data Twitter tunggal (95,97%), namun tetap bermanfaat sebagai sinyal prediktif arah harga melalui XGBoost dengan akurasi 61,11%, jauh di atas baseline acak 33%. Penerapan LoRA terbukti efisien, hanya melatih 0,51% dari total parameter FinRoBERTa tanpa mengorbankan akurasi. Penelitian ini menyimpulkan bahwa integrasi sentimen multimodal dan optimasi LoRA mampu menghasilkan sinyal sentimen yang akurat dan efisien secara komputasi untuk mendukung prediksi arah harga Bitcoin. Kata Kunci : FinRoBERTa, LoRA, analisis sentimen multimodal, Xgboost, arah harga Bitcoin.
Actions (login required)
 |
View Item |
