Performance of Optimization Algorithms in Attention-Based Deep Learning Model for Fake News Detection System

Abstract

Automatic fake news detection for categorizing the news as either fake or real is a complicated problem. Generally, the fake news does not fully contain false information. It is usually mixed with a substantial portion of genuine information. The accessibility to the Internet and willingness to distribute the data through social media is quick and easy. This makes it very easy to propagate fake news worldwide leading to dangerous and obnoxious impacts on society. Most of the current methods to tackle false news detection are based on deep learning approaches. But, these fake detection approaches did not exhibit remarkable improvement in identifying the fallacy because of the insufficiency of datasets. So, LIAR which is an efficient and benchmark dataset that is publicly available for carrying out research on fake news detection and has been used in the paper for detection of fake news. In this paper, CNN-based deep learning neural network model with attention mechanism has been considered for fake news detection system and the performance of the attention mechanism with seven different training optimization algorithms Stochastic, Adam, Nadam, Adamax, etc., has been evaluated and compared. Performance evaluation has been carried out in terms of accuracy, precision, recall, and f1-score with LIAR dataset.