Novel Trio-Neural Network towards Detecting Fake News on Social Media

Abstract

In recent days most people are using the internet to know the latest news faster, parallel false information also spreads for many reasons. The fake news is artificially manipulated and elongated by the true information, this creates negativity and diverse the users in particular opinions. Fake news detection is a more complicated and labor-consuming process because the data has kept on growing as big data. The detection of fake news using a single parameter has become less reliable and so there is a need to use multiple parameters to improve the reliability of the model. The parameters such as text, audio, video, and time were traditionally for fake news detection. In this article, the proposed model is designed to work with three parameters namely geolocation, text feed, and image data of the user in their handy smart mobile phone. The proposed Novel Trio-Neural Network has a binary classifier to detect fake or real news, the location spoofing is avoided by checking the movement probability of the user using Bayesian Geolocation Timestamp, the text feed posted by the users is analyzed by using BERT Fact Checker, and image shared by the user on the internet are mapped to text with similarity checker extracted feature from the image using VGG16 Similarity Mapping. The integrated Novel Trio-Neural Network was trained, tested, and validated with the FakeNewsNet dataset. The proposed model reached the F1-Score of 82.31%, and the performance of the model has significantly improved by 4.01% from the existing model.