In the age of information technology, although it is convenient to obtain the information of integrated media, the hidden value is difficult to excavate directly, and it is necessary to use knowledge graph to carry out intelligent analysis and interpretation. However, the construction of a knowledge graph for integrated media is time-consuming and laborious. Therefore, this paper initiates the construction of an integrated media knowledge graph for news recommendation. Firstly, this paper carried out the construction of news knowledge graph ontology. Subsequently, this paper uses large language models (LLMs) for event extraction and event relation extraction work, while this paper uses a BERT + Softmax model to predict entity-level relation within events. The automated construction from news text and related images to knowledge graph is realized. Experiments show that the integrated media knowledge graph generated through this automated construction process is of high quality, demonstrating its potential in personalized news recommendation.