MedCPT: Contrastive Pre-trained Transformers with large-scale PubMed search logs for zero-shot biomedical information retrieval

Key Points

• The aim is to enhance biomedical information retrieval by leveraging contrastive pre-trained transformers.
• Developed MedCPT model using large-scale PubMed search logs.
• Implemented zero-shot learning techniques for information retrieval.
• Made the code and model publicly available on GitHub.
• MedCPT demonstrated significant improvements in retrieval accuracy compared to baseline models.
• The system effectively retrieved relevant biomedical literature with no prior training on specific queries.