The core content of this paper is to analyze the integration of cross-media data and artificial intelligence. The core of the whole paper is to analyze different types of media data such as text, image and video. With the increasing complexity and quantity of multimedia data, it can be seen that traditional methods can no longer meet the current data needs. Therefore, some advanced technologies need to be integrated, such as convolutional neural network (CNN) Long and short memory network (LSTM) and graph neural network (GNN) to extract the data content. Therefore, the core of this paper emphasizes the centralized extraction of innocuous complex data through mixed and multi-transport facilities, and proposes that the framework can enhance information extraction and value mining, and this method can be more applied to the media medical and security fields.