EMKG: Efficient Matchings for Knowledge Graph Integration in Stance Detection

Knowledge graph (KG) has been used to provide additional relational knowledge about stance targets, thereby enhancing stance detection performance. However, the utilization efficiency of KG remains unconcerned and we find that the knowledge adoption amount is low in practice. In this work, we propose an efficient framework to improve the exploitation of KG, in which we integrate three different matching mechanisms to improve knowledge usage. A knowledge injection module is also introduced to explore enriched knowledge by constructing a supplementary set. These supplements contribute to stance detection through a text-pair classification mechanism. The experimental results on different datasets demonstrate that our approach outperforms existing works on knowledge graph exploitation and stance detection. Our code is available at https://github.com/kxy-cheng/EMKG.