The field of artificial intelligence is witnessing significant developments in the integration of language models with knowledge graphs, enabling more accurate and informative text-to-image generation and knowledge retrieval. Researchers are exploring innovative approaches to refine prompts and combine language models with knowledge graphs to overcome limitations such as hallucination and lack of source traceability. Notably, the use of hierarchical graph representation learning and ensemble learning methods is improving the prediction of associations between ingredients and diseases in traditional Chinese medicine. Furthermore, the development of domain-specific knowledge graphs and Graph-based Retrieval-Augmented Generation (GraphRAG) is enhancing the accuracy and reliability of language models in various applications. Noteworthy papers include: TextTIGER, which improves image generation performance by augmenting knowledge on entities included in prompts, and OpenTCM, which achieves high-fidelity ingredient knowledge retrieval and diagnostic question-answering in traditional Chinese medicine. Additionally, Node2Vec-DGI-EL is a hierarchical graph representation learning model that accurately predicts ingredient-disease associations.