Popularity Increment Forecast Based on Depth Fully Extracting Spatiotemporal Structure

The data characteristics of the information cascade are essential to the prediction performance of the information cascade. The traditional prediction methods can not well couple the various attributes of cascade data. This paper proposes an approach based on deep extraction of full spatiotemporal structure (FuSTCas). FuSTCas uses the deep learning framework to extract and refine the spatiotemporal information in the raw cascade data. Then FuSTCas combines the time decay function and multi-layer perceptron to make the final cascade prediction. The experimental results show that compared with several baseline models, FuSTCas does improve the prediction accuracy. And the components of FuSTCas effectively extract the spatiotemporal information.