UTILIZATION OF INDOBERT FOR SENTIMENT CLASSIFICATION OF STOCK NEWS AND CORRELATION ANALYSIS WITH PRICE MOVEMENTS ON THE INDONESIA STOCK EXCHANGE

• Main Author: Naufal Attar, Rava
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/82436
• Institution: Institut Teknologi Bandung

This study aims to classify stock news sentiment using the IndoBERT model and analyze its correlation with stock price movements on the Indonesia Stock Exchange. The data used was collected through web scraping from news portals such as Emiten News, Liputan6, and Detik.com, comprising 2,857 stock news articles. After undergoing labeling and preprocessing, the data was used to train various variants of the IndoBERT model. Experimental results show that the IndoBERT-large-p1 model provides the best performance with an accuracy of 0.78, followed by the IndoBERT-lite-large-p1, IndoBERT-base-p1, and IndoBERT-lite-base-p1 models. In the correlation analysis between stock news sentiment and stock price movements, the sector with the highest correlation was the basic materials sector, with a hit rate of 0.60 and a Pearson correlation of 0.389. This study concludes that the IndoBERT model can be effectively used for stock news sentiment analysis and demonstrates a correlation between news sentiment and stock price movements in the Indonesian stock market. However, the limited amount of data for each stock sector is a constraint in achieving higher correlation values.