Archaeological sites pose significant challenges in terms of accessibility and data integration. The Semantic Web, particularly through the CIDOC CRM ontology, enables structured data sharing. However, manually constructing knowledge graphs remains a complex and labor-intensive task. This study explores semi-automatic methods leveraging large language models (LLMs) to generate knowledge graphs for archaeological data. We propose different prompting strategies based on varying levels of information about CIDOC CRM and experiment with diverse prompt patterns. Experiments conducted on real datasets demonstrate that providing LLMs with a carefully selected subset of the CIDOC CRM ontology, combined with few-shot prompt patterns, enhances RDF extraction and improves performance in answering competency questions.