Peeking strategy for online news diffusion prediction via machine learning

For computational social scientists, cascade size prediction and fake news detection are two primary problems in news diffusion or computational communication research. Previous studies predict news diffusion via peeking the social process (temporal structure) data in the initial stage, which is summarized as Peeking strategy. However, the accuracy of Peeking strategy for cascade size prediction still should be improved, and the advantage or limitation of Peeking strategy for fake news detection has not been fully investigated. To predict cascade size and detect fake news, we adopt Peeking strategy based on well-known machine learning algorithms. Our results show that Peeking strategy can effectively improve the accuracy of cascade size prediction. Meanwhile, we can peek into a smaller time window to achieve a higher performance in predicting the cascade size compared with previous methods. Nevertheless, we find that Peeking strategy with network structures fails in significantly improving the performance of fake news detection. Finally, we argue that cascade structure properties can aid in prediction of cascade size, but not for the fake news detection.