Sentiment Effect on Mandiri Bank Stock using Feature Expansion Word2vec and Classification Method Gradient Boosted Decision Tree

In June 2019, there was an accident to the Mandiri bank system. On the same day of the accident, Mandiri bank stock price is decreasing. Many customers use social media as a means of conveying their opinions on this accident, one of the social media used is twitter. Twitter is a service used by users to upload information in the form of text that contains sentiments from its users. However, Twitter itself has a character limit, which is only 280 characters. As a result of character restrictions, opinions expressed may contain vocabulary mismatches because they contain slang, abbreviations, emoji, etc. Because of this, it is difficult to understand the context of the opinions expressed. Therefore, Word2vec feature expansion is used to deal with the problem of vocabulary mismatch, which can result in word context. This study develops a Twitter sentiment classification system using the Word2vec feature expansion method with the Gradient Boost Decision Tree classification algorithm. The results of this study indicate that by using feature expansion in the classification algorithm, model performance has increases to 3.29% for accuracy, 3.35% for precision, 3.31% for recall, and 3.31% for f1-score. From the sentiment classification model built to predict stock price movements, sentiment has no correlation towards stock price movements, with the correlation value of 0.097 for positive sentiment and 0.124 for negative sentiment.