With the development of the Internet, the web has gradually become the primary channel for people to access news. However, faced with a massive amount of content, retrieving the news of interest has become challenging. Existing news recommendation frameworks have limitations, such as insufficient information encoding for candidate news and neglecting the semantic correlation between target news and the overall news landscape. To address these issues, we suggest CDIG, a dual interactive graph model comprising a news graph and a user graph. In the news graph, we augment semantic relationships between news articles by employing a news semantic graph and leverage an advanced BGE model for representation extraction. In the user graph, we utilize a heterogeneous graph neural network and establish three types of edges: user-news, user-topic, and news-topic. We employ the HAN heterogeneous graph neural network for modeling. Finally, during the process of dual-graph interaction, we learn the representations and functionalities of both the news and user graphs. Experimental results show that CDIG surpasses traditional recommendation algorithms in performance on the MIND news dataset.