You are working with a dataset for a RAG system that generates solutions from troubleshooting guides and user manuals. During embedding, similar issues are mapped to different vectors, causing inconsistent retrieval. What is the most effective strategy to address this?