In order to extract the relevant features of rumors effectively, this paper proposes a novel rumor detection model with bidirectional graph attention network on the basis of constructing a directed graph, named P-BiGAT. Firstly, this model builds the propagation tree and diffusion tree through the tweet comment and reposting relationship. Secondly, the improved graph attention network (GAT) is used to extract the propagation feature and the diffusion feature through two different directions, and the multihead attention mechanism is used to extract the semantic information of the source tweet. Finally, the propagation feature, diffusion feature, and semantic information representation of the source tweet are connected together through a fully connected layer, and the mapping function is used to determine the authenticity of the information. In addition, this paper also proposes a new node update method and applies it to the model in order to select neighbor node information effectively. Specifically, it can select the neighbor information node with larger weight to update the node according to the weight of the neighbor node. The results of the experiment show that the model is better than the baseline method of comparison in accuracy, precision, recall, and F1 measure on the public datasets.
The higher-order and temporal characteristics of tweet sequences are often ignored in the field of rumor detection. In this paper, a new rumor detection method (T-BiGAT) is proposed to capture the temporal features between tweets by combining a graph attention network (GAT) and gated recurrent neural network (GRU). First, timestamps are calculated for each tweet within the same event. On the premise of the same timestamp, two different propagation subgraphs are constructed according to the response relationship between tweets. Then, GRU is used to capture intralayer dependencies between sibling nodes in the subtree; global features of each subtree are extracted using an improved GAT. Furthermore, GRU is reused to capture the temporal dependencies of individual subgraphs at different timestamps. Finally, weights are assigned to the global feature vectors of different timestamp subtrees for aggregation, and a mapping function is used to classify the aggregated vectors.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.