Data catalogs play an essential role in supporting efficient data discovery, integration, and reuse. Unfortunately, many of these real-world data catalogs are often sparse and lack sufficient or consistent metadata, reducing their overall utility. Existing approaches for the automatic generation of metadata, such as rule-based systems, large language models (LLMs), retrieval-augmented generation (RAG) frameworks, and knowledge graph techniques, have their own set of limitations. These models often struggle with incomplete or diverse catalogs, can produce errors or even hallucinations, and typically depend on structured input or significant human intervention to function properly. However, their performance may be further constrained by challenges such as limited coverage, scalability, and integration complexity. An Enhanced Knowledge-Augmented Metadata Generation (EKAMG) framework has been proposed to address this issue. It uses machine learning, semantic enrichment, and external knowledge bases to automatically produce high-quality metadata from sparse data sets. The suggested approach is assessed using benchmark and real-world catalog datasets and compared with several current cutting-edge metadata generation models. The EKAMG framework achieves an improvement of 25–30% in metadata completeness, 20% increase in semantic alignment accuracy, and reduces hallucination errors by almost 10% compared to existing methods.