Feature Expansion with FastText on Topic Classification Using the Gradient Boosted Decision Tree on Twitter

Twitter is a microblogging website that allows users to post and read short messages (tweets) with a character limit of 280 characters. It often leads to very short tweets, does not always utilize proper grammar, and frequently includes slang. Due to these characteristics, topic classification experiments in tweets frequently have low accuracy. As a result, feature expansion is used in this study to decrease word mismatches. Some researchers have previously conducted research with feature expansion using Word2Vec, WordNet, etc. However, as far as researchers are looking, no one has used fastText for feature expansion, therefore this study aims to identify the effect of feature expansion with fastText when paired with the Gradient Boosted Decision Tree algorithm. The results showed that the topic classification system using the Gradient Boosted Decision Tree method and with feature expansion using fastText resulted in the highest accuracy of 91.39%.