BERT-BiLSTM: A Novel Hybrid Approach for Stance Detection on Political Tweets

Stance detection has become increasingly important in modern society as people actively express their opinions on social media platforms, particularly on Twitter. Despite significant advancements in stance classification using machine learning and deep learning techniques, challenges remain due to the complexity of language and the inherent noise in user comments. Although significant research work has been done in this area, a prominent shortcoming lies in the ability to deal with contextual comprehension and sequential dependencies, which are critical for precisely identifying implicit or indirect stances. To tackle this challenge, the paper presents a hybrid approach known as BERT-BiLSTM by fine-tuning the BERT model on political tweets. The proposed model’s performance is evaluated on two benchmark datasets, classifying the stance into three distinct categories: Against, Favor, and None. By integrating the advantages of BERT’s contextual sensitivity with LSTM’s capability to maintain the chronological order of information, this hybrid model markedly enhances accuracy in stance detection within political discourse, achieving an accuracy of 79.60% and 56.13% on the P-stance and SRQ datasets, respectively.