A BERT-BiLSTM Approach for Socio-political News Detection

The detection of socio-political news is critical for ensuring informed citizens, holding media accountable, understanding public sentiment, detecting misinformation, guiding policy development, monitoring crises, facilitating international relations, and ensuring public safety. In this paper, a novel model is proposed as the BERT-based BiLSTM (Bidirectional Encoder Representations from Transformers–Bidirectional Long Short-Term Memory) model to balance the accuracy with computational efficiency and detect socio-political news. To begin, the BERT model is used instead of the traditional word embedding model to represent the word vector. The text representation is calculated based on the contextual information, and it is adjusted according to the sentiment of the word while the contextual information is fused. Second, the BiLSTM model is linked to the BERT model, allowing the BiLSTM model to extract socio-political news features from both directions at the same time. The proposed model’s performance is evaluated and compared with stats of art methods. In the comparison, the BERT-BiLSTM model detects socio-political news with an accuracy of 96.8%.