Hoax Detection on Indonesian Tweets using Naïve Bayes Classifier with TF-IDF

Abstract

Twitter is one of the most popular social media platforms in the world nowadays. Twitter users in Indonesia are the fifth largest in the world and are always active in expressing themselves and getting information through tweets. A hoax is a lie created as if it were true. Hoaxes are also often spread via tweets. The spread of hoaxes is extremely dangerous because it can cause social discord and even misunderstanding. Therefore, hoaxes must be resisted. This study aims to build a system to detect hoaxes on Indonesian tweets. The objective of this research is to identify hoax Indonesian tweets by using the Naïve Bayes classifier with Term Frequency Inverse Document Frequency (TF-IDF). This study collects and annotates tweets from hoax tweets post which sent by a user account. This study also applied several text preprocessing techniques to provide datasets. To provide the best hoax prediction model, this work splits datasets into training and testing datasets. There are four experimental scenarios that refer to splitting the dataset. The experimental results showed that the hoax prediction model using Naïve Bayes with TF-IDF had 64% accuracy and recall, 69% and 67% precision, and a F1-score respectively. This result is also superior to the hoax prediction model when using the Naïve Bayes classifier without the TF-IDF. It means that TF-IDF has made a positive contribution to improving model performance. Finally, this research contributes by detecting news with a proclivity for hoaxes and filtering what is classified as hoaxes or not.