RoBERTa-BiLSTM: A Chinese Stance Detection Model
Qiumei Pu, Fude Li · 2024
In this paper, we propose a text classification model for the problem of stance detection in Chinese texts, based on the RoBERTa pre-trained language model and the Bidirectional Long Short-Term Memory Network (BiLSTM). The purpose of stance detection is to identify the author's attitude towards a specific topic, such as support, opposition, or neutrality. Our model combines the deep semantic understanding capabilities of RoBERTa with the sequence information capturing ability of BiLSTM to improve the accuracy of Chinese text stance detection. Experiments conducted on the NLPCC-2016 Task4 Chinese Weibo Stance Detection dataset demonstrate the effectiveness of our method. The experimental results show that our model significantly improves accuracy and other key performance metrics, indicating that combining RoBERTa and BiLSTM is an effective strategy for addressing Chinese stance detection tasks.