|
|
Rumor detection method based on graph convolution network and Attention Mechanism
|
Fenglizhou,Liu Furong,Wang Youwei
|
(School of Statistics, Tianjin University of Finance and Economics, Tianjin 300222, China)
( School of Information, Central University of Finance and Economics, Beijing 100081, China)
|
|
|
Abstract
[Objective] A rumor detection method based on graph convolutional network and attention mechanism is proposed to address the issue of insufficient consideration of forwarding relationship features and text semantic features between comments in current rumor detection methods.
[Methods] Firstly, analyze the forwarding and reply relationship characteristics between comments and construct a comment relationship feature map, fully mining the correlation characteristics between comments. Then, according to the text semantic similarity between comments, the BERT model is used to generate the vectorization representation of sentences, and the semantic feature map of comments is constructed by calculating the cosine similarity to fully extract the semantic relevance of comments. Finally, based on Graph Convolutional Network (GCN), information transmission between different nodes was completed, and attention mechanism was used to distinguish the impact of source comments and other comments on rumor detection during the information transmission process of each node, thereby obtaining an accurate representation of the comment nodes.
[Results] Experiments were conducted on public datasets, and the results showed that the accuracy of our method in the Twitter15 and Twitter16 datasets reached 0.86 and 0.87, respectively, with F1 mean values of 0.86 and 0.87, respectively. Compared with the BiGCN method, the accuracy was improved by 5.1% and 1.5% on the Twitter15 and Twitter16 datasets, respectively, and the F1 mean was improved by 5.3% and 2.3%, respectively.
[Limitation] This method only uses text data for rumor detection. In future research, it will consider combining features such as images, user attributes, and time attributes to further improve the accuracy of the model.
[Conclusion] The application on public datasets verifies that the proposed method can effectively improve the performance of rumor detection, providing valuable reference for rumor recognition and detection tasks.
|
Published: 08 January 2024
|
|
|
Viewed |
|
|
|
Full text
|
|
|
|
|
Abstract
|
|
|
|
|
Cited |
|
|
|
|
|
Shared |
|
|
|
|
|
Discussed |
|
|
|
|