The growth of social media has greatly promoted the dissemination of real-time information among users, and also provided a breeding ground for the spread of fake news. User comments on social media platforms contain rich emotional reactions and semantic information, which provide important clues for fake news detection. However, most existing fake news detection methods primarily focus on semantic information in both content and comments, ignoring emotional information and the mutual selection relationship between content and comments. To address these limitations, we propose a novel ConComFND model. This model initially extracts emotional features from both news content and user comments, followed by the utilization of Text-CNN to capture semantic features. Furthermore, we introduce a content-comment cross-attention mechanism to fuse information selectively, thereby enabling the model to focus on more relevant information. The concatenation of these extracted features is finally employed for fake news detection. Extensive experiments conducted on both Chinese and English datasets reveal that the proposed ConComFND model significantly improves detection accuracy, achieving an enhancement of up to 8.6% compared to traditional HAN-based approaches.