Hybrid Deep Learning Model for Rumor Detection in Social Media

Abstract

This paper presents a hybrid deep learning approach for rumor detection in social media that combines transformer-based text representation with graph-based modeling of information propagation. The proposed method integrates a Bidirectional Encoder Representations from Transformers model for extracting semantic features from textual data and a Graph Attention Network for capturing structural relationships between messages in conversation threads. This enables the model to effectively utilize both content and propagation patterns for improved classification performance. Due to the combination of textual and graph-based features, the proposed approach enhances the accuracy and robustness of rumor detection, particularly in the presence of noisy and imbalanced data. Experimental evaluation conducted on the PHEME Dataset demonstrates that the hybrid model outperforms traditional machine learning methods as well as standalone deep learning models, achieving higher Precision, Recall, and F1-scores. The proposed approach provides an effective solution for detecting misinformation in social media environments and can be extended to other domains involving structured and unstructured data.