Sentiment Analysis of Indonesian Economics News Summary on The ICI using Long Short Term Memory

Abstract

Many investors use news summary as a reference while deciding whether to purchase, sell, or hold shares as part of their investment activity. The information provided by the news, on the other hand, frequently fails to meet expectations. Therefore, this study aims to classify economic news summary using sentiment analysis method followed by analyzing the results to determine the correlation between Indonesian economic news summaries and Indonesia Composite Index (ICI) trends. The dataset for this study is a collection of news articles from the Kontan online news site. This study begins with the collection of a dataset that includes 1720 training data and 440 test data that have been automatically summarized using the cosine similarity approach and labeled (positive/negative) based on the ICI trend the next day after the data was crawled. We use Term Frequency – Inverse Document Frequency (TF-IDF) as the word weighting technique and Long Short Term Memory (LSTM) method as the classifier to build the model. From the model training process that has been carried out for eight trials by tuning hyperparameters, the training accuracy is 86% and the F1-Score is 89.3%. While the testing accuracy and F1-score are 68.1% and 75.8% consecutively