Knowledge graphs (KGs) form the foundation of artificial intelligence, as they provide rich, semantically meaningful, and structured descriptions of world entities and their relationships. They add data connectivity, reasonability, and machine comprehension to various fields of study. The article introduces new knowledge graph (KG) frameworks and deployment methods, ranging from manual curation to automatic extraction, as well as approaches that combine both paradigms. Their integration with large language models (LLMs) is investigated, and their contribution to knowledge representation, entity resolution, and context-aware retrieval is exemplified. KGs are implemented in various applications such as natural language processing, healthcare, finance, cybersecurity, e-commerce, and education. Their greater usage provides better scalability, dynamic updateability, and effortless interoperability between diverse datasets. KGs are affected despite their superiority with data sparsity, computational complexity, standardization, and multimodal data integration. Available benchmarks highlight how the performance of models such as GPT-4 can support RDF-based KGs, but even they find it difficult to deal with compound SPARQL queries effectively. Future research directions include enhancing multimodal knowledge fusion, query execution optimization, and self-adapting knowledge graphs. Resolution of these issues will necessitate cooperation among AI researchers, data scientists, and knowledge engineers. The current paper gives a synopsis of knowledge graphs (KGs), their changing role in AI applications, and how they dynamically evolve. It also thoroughly reviews diverse methodologies, frameworks, and the latest research achievements in knowledge graph studies during the last three years.