A Parallel Neural Network Approach for Faster Rumor Identification in Online Social Networks

The unprecedented scale of rumor propagation in online social networks urges the necessity of faster rumor identification and control. The identification of rumors in the inception itself is imperative to bring down the harm it could cause to the society at large. But, the available information regarding rumors in inception stages is minimal. To identify rumors with data sparsity, we have proposed a twofold convolutional neural network approach with a new activation function which generalizes faster with higher accuracy. The proposed approach utilizes prominent features such as temporal and content for the classification. This rumor detection method is compared with the state-of-the-art rumor detection approaches and results prove the proposed method identifies rumor earlier than other approaches. Using this approach, the detected rumors with 88% accuracy and 92% precision for experimental datasets is 5% to 35% better than the existing approaches. This automated approach provides better results for larger and scale-free networks.