Hoax News Detection Using Passive Aggressive Classifier And TfidfVectorizer

Abstract

Indonesia is one of the countries with the highest number of social media users. Million social media users in Indonesia reached 167 million in January 2023. These users are spread, across various social media, including Twitter with 24 million users. The high number of social media users on Twitter makes the information validation process even more neglected. Moreover, the trend of news interest read by social media users is only adjusted to their individual tastes. This phenomenon is evidenced by the large number of fake news (hoaxes) circulating in society which are spread through social media. Therefore, an accurate machine learning model is needed to classify "real" and "hoax" news. This study uses the TfidfVectorizer algorithm and Passive Aggressive Classifier for datasets that are shared through the Kaggle site. The contents of the dataset were sourced via social media Twitter over a span of 5 years, namely 2015-2020. At the preprocessing stage to making the Confusion Matrix, the machine learning model shows that it can work well as expected, namely getting Accuracy, Precision, and Recall scores of 82.44%, 80.66%, and 82.44%. In addition, the results of the confusion matrix show that in the dataset used, there is more "real" news than "hoaxes", that is, the model is able to predict 1059 real news and 211 hoax news, with actual conditions 1106 real news and 164 hoax news.