Using Convolution Neural Network with BERT for Stance Detection in Vietnamese

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 7220–7225 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 7220 Using Convolution Neural Network with BERT for Stance Detection in Vietnamese Oanh Thi Tran*, Anh Cong Phung**, Bach Xuan Ngo** *International School, Vietnam National University, Hanoi **Posts and Telecommunications Institute of Technology, Vietnam oanhtt@isvnu.vn, anhpc@ptit.edu.vn, bachnx@ptit.edu.vn Abstract Stance detection is the task of automatically eliciting stance information towards a specific claim made by a primary author. While most studies have been done for high-resource languages, this work is dedicated to a low-resource language, namely Vietnamese. In this paper, we propose an architecture using transformers to detect stances in Vietnamese claims. This architecture exploits BERT to extract contextual word embeddings instead of using traditional word2vec models. Then, these embeddings are fed into CNN networks to extract local features to train the stance detection model. We performed extensive comparison experiments to show the effectiveness of the proposed method on a public dataset. Experimental results show that this proposed model outperforms the previous methods by a large margin. It yielded an accuracy score of 75.57% averaged on four labels. This sets a new SOTA result for future research on this interesting problem in Vietnamese. Keywords: stance detection, Vietnamese, BERT, CNN 1. Introduction With the development of information technology, peo- ple nowadays can easily express their stances against specific topics on social media like sites, news portals, forums, and online newspapers. Stance (Dilek and Fa- zli, 2020) is defined as the expression of the speaker’s attitude, standpoint, and judgement toward a target. The analysis of these comments is typically used as an initial step for fake news detection (Sun et al., 2018) or in understanding public reaction towards the social issues mentioned in these claims. Stance detection is a newly emerging research field with the goal is to iden- tify the stance of the text authors toward a target which is either explicitly mentioned or implied within the text. An example is given in Figure 1, which shows a head- line of a news article and its four comments. This arti- cle received lots of the follow-up comments which in- dicate the reaction of public users (i.e. agree, disagree, discuss or unrelated) towards the claim mentioned. To deal the task, researchers framed it as a classifica- tion problem, and then exploit different machine learn- ing (ML) methods to solve it. So far, there existed many works proposed to solve the task using features- based ML approaches (Kucher et al., 2018; Simaki et al., 2017; Swami et al., 2017). At present, deep learn- ing approaches (Chung et al., 2014; Sun et al., 2018; Zhang et al., 2017) have been extensively investigated and yielded better performance for this task. There were also many datasets in popular languages made publicly available for research community via official competitions or challenges like SemEval-2016 Task 6: Detecting Stance in Tweets1, Shared Task of Stance Detection in Chines Microblogs at NLPCC- 1https://alt.qcri.org/semeval2016/task6/ ICCPOL-20162, …

Content truncated to 500 words