Fake News Detection Using Content-Based Features and Machine Learning

摘要

The problem of fake news is a complex problem and is accompanied with social and economic ramifications. Targeted individuals and entities may lose trustworthiness, credibility and ultimately, suffer from reputation damages to their brand. Economically, an individual or brand may see fluctuations in revenue streams. In addition, the complex nature of the human language makes the problem of fake news a complex problem to solve for currently available computational remedies. The fight against the spread of fake news is a multi-disciplinary effort that will require research, collaboration and rapid development of tools and paradigms aimed at understanding and combating false information dissemination. This study explores fake news detection techniques using machine learning technology. Using a feature set which captures article structure, readability, and the similarity between the title and body, we show such features can deliver promising results. In the experiment, we select 6 machine learning algorithms, namely, AdaBoost as AB, Decision Tree as DT, K-Nearest Neighbour as KNN, Random Forest as RF, Support Vector Machine as SVM and XGBoost as XGB. To quantify a classifier’s performance, we use the confusion matrix model and other performance metrics. Given the structure of the experiment, we show the Support Vector Machine classifier provided the best overall results.