Providing a Hybrid Approach Based on Deep learning and Machine Learning to Detect Fake News - A Case Study of Persian Newsin the Field of COVID-19

False or unconfirmed information is published on the web like accurate information, so it can become viral and influence public opinion and decisions. Fake news and gossip show the most popular forms of false and unverified information, respectively, and they should be detected as soon as possible to avoid significant effects Interest in effective identification techniques has been increasing in recent years.The problem of detecting fake news is known as a classification problem in natural language processing and text mining, and its purpose is to distinguish fake news from real and extracted texts, and to improve the accuracy of detecting fake news is the main issue of this research. Convolutional neural networks, as one of the most important models of deep learning, have gained high accuracy on these issues. These networks include problems such as not considering the position of words, which is solved by using the capsule network, and in order to achieve optimal accuracy, two problems of heavy processing of all connected layers and reducing the parametric space using the algorithm XGBOOST and particle swarm optimization (PSO) algorithm are proposed.

This study is an applied research in which about 42,000 Persian news from different cities of Iran were collected from Twitter and using additional methods of cleaning and preprocessing, additional information was removed and after tagging, the news was ready to be used for the proposed approach using Python software and related libraries are equipped with machine learning and deep learning algorithms.

During testing, some machine learning algorithms had more power in classification problems, but with the changes in the structure of the convolutional network and Capsul network algorithm, better results were obtained than machine learning algorithms and other similar algorithms.

The proposed solutions in this research in comparison with the approaches of basic algorithms or solutions to solve the mentioned problems by replacing the optimal classifier and reducing the parametric space, by changing the input has been able to achieve better and more acceptable results than other approaches. And achieve an accuracy of about 96%.