We are happy to present an article accepted for the CSEDU (17th International Conference on Computer Supported Education), which will take place April 1-3 in Porto, Portugal.
A Collaboration to Integrate Cutting-Edge Learning Technologies
SuperMemo, a pioneer in spaced repetition learning, and Medico, a comprehensive digital resource for Polish doctors, are excited to announce a collaborative project aimed at transforming how medical students and residents in Poland prepare for their State Specialization Examination (PES). This initiative integrates the strengths of both platforms to provide an innovative and effective learning experience.
Medico, owned by Polish Scientific Publishers PWN (Wydawnictwo Naukowe PWN), offers an extensive medical knowledge base with over 120,000 documents, including textbooks, journal articles, clinical guidelines, and more. Its advanced search engine enables users to quickly retrieve precise information on specific clinical issues. SuperMemo, on the other hand, is renowned for its spaced repetition algorithms, which optimize memory retention through systematic review scheduling. This method predicts forgetting curves for individual learners and information, prompting reviews at optimal times to minimize repetitions while maximizing retention.
The core of this collaboration lies in creating PES preparation courses that leverage Retrieval-Augmented Generation (RAG) to generate insightful comments for nearly 20,000 questions, as of February 2025. The State Specialization Examination (PES) is a critical assessment for physicians and dentists in Poland seeking specialist certification. These courses are designed to help aspiring specialists master the knowledge and competencies required for success.
RAG: Enhancing Learning with Contextual Insights
RAG is an innovative approach that combines information retrieval with text generation. In this context, it means that for each PES question, the system retrieves relevant excerpts from Medico’s vast database of medical content and uses these excerpts to generate a comprehensive comment explaining the correct answer and providing additional context. This ensures that learners not only know the answer but also understand the underlying principles and rationale.
The quality of these RAG-generated comments is paramount to the effectiveness of the learning experience. The research described in the source article focuses specifically on enhancing the relevance of the source documents used as context for RAG. The more relevant the source material, the more accurate, credible, and coherent the generated comments will be. This is why a significant portion of the project was dedicated to refining the retrieval system, query rephrasing, and reranking mechanisms.
Significant Improvements in Document Relevance
The collaboration has significantly improved document relevance through rigorous experimentation and evaluation by medical annotators. The number of relevant documents increased from 4.59 to 6.83 out of 10. This enhancement ensures that the RAG system provides learners with the most pertinent and reliable information, minimizing cognitive overload and maximizing knowledge retention.
This integration promises to deliver a scalable, high-quality, and individualized educational resource that addresses the unique needs of medical education in Poland. The team expects the Medico-SuperMemo integration to be used in practice starting in March 2025.
Read the full article “Optimizing Retrieval-Augmented Generation of Medical Content for Spaced Repetition Learning”
