Knowledge Graph (KG) is a technology that employs graph models to represent knowledge and model the interconnected relationships among entities. The purpose of knowledge graph completion techniques is to automatically fill in missing entities, attributes, and relationships within an existing knowledge graph through inference and learning methods. In recent years, neural network models have emerged as powerful learning frameworks and have found extensive applications in the domain of knowledge graph completion. This paper presents a comprehensive investigation and analysis of the application of neural network models in knowledge graph completion techniques. Explored the challenges faced by knowledge graph completion techniques such as sparse data. Point out the future development direction, multimodal fusion and incremental learning knowledge graph completion. The objective is to provide valuable insights and assistance for the advancement of research in knowledge graph completion techniques.