.png){kind=icon}
.png){kind=icon}
Return
DR-DGRNet: A Model for Intent-Based Disinformation Recognition Using Dynamic Graph Representation Learning
.png){kind=icon}
PRE
AI
DOI：10.1109/TCSS.2026.3654553.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The rapid development of social platforms has enabled users to easily manipulate machine-controlled spammer accounts powered by large language models (LLMs) to spread misinformation for ulterior motives. To address this, it is crucial to model the dissemination space of intent-based events and detect them early. This article proposes a model for recognizing intent-based disinformation using dynamic graph representation learning, named DR-DGR. First, a group behavior quantification algorithm based on the attention mechanism is introduced. This algorithm incorporates relative position encoding into the attention mechanism to extract subspace user group features. Second, an algorithm for user behavior representation is proposed. Considering the importance of both group and individual behaviors in the subspace, convolutional and attention-based components are employed to mine individual and group behavioral features, which are then fused to represent the dissemination subspace. Finally, to account for the varying importance of features over time, key dissemination features are dynamically extracted using an attention component. Moreover, to mitigate early-stage data sparsity, source content is incorporated to enhance early recognition performance. Evaluations on large-scale public datasets and real-world environments demonstrate that DR-DGR effectively identifies disinformation.
Keywords:
Attention mechanism
disinformation recognition
dynamic fusion
space representation
Journal
IF:
4.9
Papers:
577
Citations:
6.8K
View more organizations