Efficient fine-tuning methodology of text embedding models for information retrieval: contrastive learning penalty (clp)