Infusing external knowledge into user stance detection in social platforms.

This article focuses on a novel stance detection model called KRGGNN, which enhances the classification of user reviews on social platforms by integrating external knowledge from knowledge graphs and structural information between reviews. The model maps keywords from reviews to the WordNet knowledge graph to extract relevant contextual subgraphs, filters noise using personalized PageRank (PPR), and encodes these subgraphs with relational graph convolutional networks (RGCN). It then employs a gated graph neural network (GGNN) to capture structural relationships within conversation threads of reviews. Experimental results on a multi-topic dataset from Reddit and Twitter demonstrate that KRGGNN outperforms six benchmark models, improving macro-average F1 scores by 1.5%–6.9%, indicating that combining external knowledge and structural features effectively boosts stance detection performance. The study also discusses challenges such as data imbalance and the potential for future work incorporating temporal modeling and bias reduction techniques.