An Approach to Identify Epidemic Diseases Rumors in Social Networks ‎using Deep Learning Techniques

One of the most important issues in social networks is the high volume of rumors that are spread by human or machine ‎agents. In such situations, automatic detection of rumors to keep public opinion safe from their potential dangers is of ‎great importance. In this research, using deep learning techniques, a new solution for automatically detecting rumors ‎related to epidemic diseases in social networks will be presented. In the proposed method, first the content of existing ‎messages is prepared for processing in the next steps. Also, weight matrix format has been used to describe content ‎characteristics. Then, in the second step of the proposed method, the convolutional neural network is used to extract the ‎set of suitable features from the matrix of features obtained from the previous step. In this way, the matrix of content ‎features is used as the input of the deep neural network, and the weight values obtained in the last fully connected layer of ‎this neural network are used as the features extracted from it. Finally, the aggregation of several binary classifiers is used ‎in order to detect rumors and classify the features extracted through convolutional neural network. For this purpose, the ‎extracted features are simultaneously processed by several learning models and the final output of the proposed system is ‎determined by voting the outputs of these three algorithms. The results of this research show that by using the proposed ‎method, rumors can be detected with an average accuracy of 98.8%, which shows an improvement of at least 2.4% in ‎detection accuracy compared to the previous methods.‎