Design and Implementation of a Web-Based Fake News Detection System Using Machine Learning

In light of the exponential increase of misinformation on the Internet, there is a need for more efficient automatic methods for its detection. In this paper we propose a Web-based Fake News Detection System that combines machine learning and NLP techniques to classify news articles instantly. Our proposed architecture is based on the combination of several supervised learning techniques, namely Naïve Bayes, Logistic Regression, Random Forest and Support Vector Machine (SVM), controlled by majority voting mechanism along with confidence-based tie-breaking scheme. Data is prepared using NLP preprocessing techniques and then vectorized using a hybrid approach of TF-IDF encoding, unigram and bigram features being used. In order to make the statistical classification more accurate and reliable, we incorporate multi-level fact checking into the system, verifying data by named entities through Wikipedia search engine, numeric values validity check, pattern recognition of scams and, optionally, through external Fact Check API provided by Google. Built in Python through Flask framework using the Agile approach, our system was tested on a balanced set of 11,632 labeled articles. The ensemble method reached the accuracy score of approximately 91% with much higher reliability than other classification approaches.