Cross-domain text classification aims to improve the performance of models trained on source domains when applied to target domains. However, domain specificity limits the model’s generalization ability. Domain Generalization (DG) seeks to enable models to perform well on unseen domains without requiring target-domain data. This paper proposes a novel DG framework ‘CFEO’, which focuses on the data’s causal structure. CFEO integrates Causal AutoEncoder (CAE) to extract causal features and leverages an optimization strategy Causal Invariant Learning with Supervised Contrastive Learning (CICL) to ensure feature consistency across domains. Experimental results demonstrate that CFEO outperforms existing methods, enhancing domain generalization and achieving state-of-the-art performance.