An Intelligent Framework for Automated Fake News Detection

Abstract

As digital news platforms and social media grow and multiply, misinformation has become a widespread problem and automatic methods of detecting fake news are becoming more important. This paper introduces a machine learning-based system to detect fake news by applying traditional classification techniques and Text Representation method Term Frequency-Inverse Document Frequency (TF-IDF). The experiments were carried out on the ISOT Fake and True News dataset which has 44,898 news articles including 23,481 fake news and 21,417 real news articles. The articles were pre-processed for each of the articles to convert the title and body in to lower case and then to exclude all the HTML elements, Punctuation, numbers and the stop words from the articles. TF-IDF (with unigram, bigram) was used for the representation of the resulting text. To train the classifiers, the data was split into an 80:20 ratio to develop the set and train the classifiers. The classifiers were trained by dividing the data set into training and testing sets in the ratio of 80:20. The accuracy, precision, recall, F1 score, and confusion matrix analysis were used to evaluate their performance. Among the tested models, precision, recall and F1 Score were 99.79%, 99.91% and 99.85% respectively with test accuracy of 99.86% with XGBoost. The study results reveal that TF-IDF based text representation along with ensemble learning can be a useful method to classify fake news on the chosen data set. As such, the study proposes an automated fake news detection using a machine learning framework with the adoption of traditional classification methods and ensemble techniques.