FGRCAT: A fine-grained reasoning framework through causality and adversarial training

Hanghui Guo, Shimin Di, Zhangze Chen, Changfan Pan, Chaojun Meng, Jia Zhu

Abstract: Highlights•FGRCAT enhances PLMs with causalities and transformation-based GNN reasoning.•The framework employs a dual strategy for adversarial training, boosting generalization.•Experimental results show FGRCAT achieves over 10% accuracy improvement compared to baseline models.•The framework outperforms large language models in capturing nuanced relationships between entities.