Adversarial Classification Rumor Detection based on Social Communication Networks and Time Series Features

• Published in: Proceedings / Asia Pacific Software Engineering Conference pp. 11 - 20
• Main Authors: Zhang, Xinyu, Chang, Zixin, Li, Li, Zhou, Wei, Wen, Junhao
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.12.2024
• Subjects: Adversarial Learning; Aggregates; Collaborative filtering; Communication networks; Contrastive learning; Feature extraction; Graph convolutional networks; Recurrent neural networks; Social networking (online); Software engineering; Time series analysis
• ISSN: 2640-0715
• DOI: 10.1109/APSEC65559.2024.00012

With the unprecedented power of social media and public opinion, the widespread rumors on social media have brought a severe negative impact on public life. This paper introduces a new rumor detection method STAC(Spatial-Temporal Adversarial Classifier). STAC aggregates rumor propagation at a finer granularity by jointly modeling their spatial structural and time-series features. In addition, STAC uses an auxiliary adversarial topic classifier to improve generalization to emerging news topics. Experimental results show that STAC exhibits advantages in early detection and overall performance. STAC employs a graph convolutional network to capture the structural information of the propagation graph and uses a bidirectional recurrent neural network to learn the sequence features of the propagation path. Through a gating mechanism, STAC aggregates the learned structural features with temporal features in a fine-grained manner. STAC proves its effectiveness in the experiments by showing advantages in early detection and overall performance. STAC performs better in rumor detection tasks and can detect rumors at an early stage.