multi-qa-mpnet-dot-v1-68261cc0·1 events·first seen Aliases: multi-qa-mpnet-dot-v1
Researchers propose fine-tuning Sentence Transformer models on a domain-specific corpus of 3,499 semantic pairs drawn from five European security standards to automate the mapping of cloud security controls to technical metrics. The training set is expanded to up to 13,996 samples via back-translation and LLM-based paraphrasing, and five architectures are evaluated on two tasks. Fine-tuned models consistently outperform zero-shot baselines, with the best achieving gains of up to 23 nDCG@10 points on the control-to-metric task. The study confirms that in-domain training data is the primary performance driver for this compliance automation use case.