Attention based Information Cascade Prediction Model

Abstract

Information cascade prediction is a research field in the field of social network analysis. It learns the transmission mode of information in online social media through the diffusion sequence and topology of information cascade. The existing research mainly focuses on using the method of cyclic neural network to learn the time structure information or spatial structure information of information cascade, while ignoring that there is no relationship between the expansion of cascade sequence. In order to solve this problem, this study proposes a new learning method, that is, the information cascade prediction model based on deep hierarchical attention (DeepHA-ICPM). The model learns the time structure information of the node sequence and the spatial structure information between the node representations through the cyclic neural networks of self attention mechanism and multi head attention mechanism respectively, and then jointly models the information cascade structure information through the hierarchical attention network Through the data set test, the model has achieved leading prediction results on the three data sets of HEP-PH, BlogCatalog and Flickr, and has good generalization ability.