Table 9 Comparative analysis of BERT and GNN-based approaches for fake news detection.

Singh and Jain³¹

Hybrid model combining BERT for semantic embedding with GNN to model relational dependencies among socio-political news elements

RSS Feeds

Achieved superior accuracy, precision, and recall compared to traditional ML and standalone DL models

Bhowmik, Mondal, and Arifuzzaman³²

GNN and transformer-based framework for Bangla sentiment analysis using BERT, Word2 Vec, FastText

Bangla Sentiment Dataset (15,114 samples)

Accuracy: 0.8957, Precision: 0.9056, Recall: 0.8515, F1-score: 0.8635

Zhang³³

GBCA (Graph BERT Co-Attention) integrating GCN and BERT for feature fusion

Three benchmark fake news datasets

Outperformed baselines in both accuracy and training efficiency

Proposed Study

Dual-Stream Graph-Augmented Transformer integrating BERT with GNN through semantic-guided graph fusion and co-attention

GossipCop, PolitiFact, FakeNewsNet

Accuracy: 0.99, Precision: 0.99, Recall: 0.987, F1-score: 0.98