To provide a structured way to perform human evaluation, we propose the first and most comprehensive guidance using some commonly used evaluation metrics in a tool form called HumanELY. Our approach and tool helps perform evaluation of LLM outputs in a comprehensive, consistent, measurable and comparable manner. HumanELY comprises of 5 key evaluation metrics of relevance, coverage, coherence, harm and comparison. Additional submetrics within these 5 key metrics provide for likert scale based human evaluation of LLM outputs.
Cite us: Awasthi, R., S. Mishra, D. Mahapatra, A. Khanna, K. Maheshwari, J. Cywinski, F. Papay and P. Mathur (2023). "HumanELY: Human evaluation of LLM yield, using a novel web-based evaluation tool." medRxiv: 2023.2012.2022.23300458.
Access the tool: https://humanely-3ora2i74hq-ue.a.run.app/