<![CDATA[The movement of energy sector stock prices exhibits high volatility, which is influenced by historical data and news sentiment. This research evaluates the performance of stock price prediction by integrating numerical data and economic news texts using a multimodal Long Short-Term Memory (LSTM) architecture. The data utilized includes the stock prices of ADRO, PGAS, and INDY from the 2021 to 2026 period, alongside Indonesian-language economic news. Sentiment extraction from the news texts was conducted using the IndoBERT model. The results indicate that the IndoBERT model achieved an accuracy and F1-score of 83%. The evaluation of the unimodal model (historical data only) yielded a Mean Absolute Percentage Error (MAPE) of 3.62% for ADRO, 3.17% for PGAS, and 5.90% for INDY. Meanwhile, the multimodal model, which combined numerical and sentiment features, resulted in a MAPE of 4.00% (ADRO), 5.46% (PGAS), and 8.47% (INDY). In conclusion, the unimodal LSTM model proved to be effective; however, the integration of sentiment features in the multimodal scheme did not provide a significant improvement in accuracy due to the highly volatile nature of the stocks.]]>
Copyrights © 2026
