Building A Medical Knowledge Concept Graph for Adaptive Learning Applications and Curriculum Mastery

June 5, 2017 2:00 PM – 3:30 PM

Toufeeq Ahmed, Vanderbilt University
Tao Le, ScholarRx

To build adaptive learning technology that understands learners’ current knowledge and can guide their next steps, we need to give computers capabilities to learn concepts and their connections. Humans are good at conceptualizing and creating mental models. Now, using natural language processing (NLP) and a learning concept graph from curriculum content, we can build smarter learning applications. In this innovcation demonstration (for students, faculty, application developers and academic leadership), we will showcase how we built a large concept graph (5 million nodes) using Neo4j, a publicly available graph database, that we will leverage into an adaptive learning platform.

Leveraging our NLP engine (QuickMatch), which can parse 30K sentences per minute, we parsed and extracted concept graphs of 6,407 nodes from First Aid USMLE Step 1 and 7,563 nodes from USMLE-Rx Step 1 Qmax, both of which are commercially available resources for USMLE preparation. We merged these graphs with a large graph extracted using Wikipedia dataset (5 million nodes). This concept graph represents all the medical concepts a medical student need to master for the USMLE Step 1. Similarly, we can parse and extract a concept graph for instructional content typically found in a four-year UME curriculum.