User behavior prediction model based on implicit links and multi-type rumor messages

Traditional prediction models of rumor forwarding are based solely on explicit network topology, and with no consideration for homogeneity and antagonism among multi-type rumor messages. To solve these problems, this study proposes a user behavior prediction model based on implicit links and multi-type rumor messages. First, because most existing studies are based on explicit network topology and ignore the influence of implicit links on information transmission, this study considers the interaction and similarity among users comprehensively and uses the K-dimension-tree algorithm to mine implicit links among non-friends, thereby improving the network topology. Second, given fuzziness and complexity of user forwarding behavior in multi-type rumor messages, considering the advantages of graph convolutional networks (GCNs) model in network representation, rumor information, user characteristics and network structure are fully represented with features. Finally, considering the high integration ability and adaptive ability of model fusion, a softmax layer is added to finalize the basic multi-classification, and then multiple GCN-based models are fused by a voting mechanism to realize the prediction of user forwarding behavior. Experiments show that the proposed model can effectively predict a user's forwarding behavior under multi-type rumor topics, and the model has improved generalization ability.(c) 2023 Elsevier B.V. All rights reserved.

Automated summary

Traditional prediction models of rumor forwarding based solely on explicit network topology are not effective due to the lack of consideration for homogeneity and antagonism among multi-type rumor messages. This study proposes a user behavior prediction model based on implicit links and multi-type rumor messages to address these problems. By considering user interactions and similarities comprehensively, implicit links among non-friends are mined using the K-dimension-tree algorithm to improve the network topology. The advantages of graph convolutional networks (GCNs) model in network representation are utilized to fully represent rumor information, user characteristics, and network structure, resulting in improved generalization ability of the proposed model.