This paper presents an LLM-powered approachfor generating concept maps to enhance digi-tal reading comprehension in higher education.While particularly focused on supporting neu-rodivergent students with their distinct informa-tion processing patterns, this approach benefitsall learners facing the cognitive challenges ofdigital text. We use GPT-4o-mini to extractconcepts and relationships from educationaltexts across ten diverse disciplines using open-domain prompts without predefined categoriesor relation types, enabling discipline-agnosticextraction. Section-level processing achievedhigher precision (83.62%) or concept extrac-tion while paragraph-level processing demon-strated superior recall (74.51%) in identify-ing educationally relevant concepts. We im-plemented an interactive web-based visualiza-tion tool https://simplified-cognitext.streamlit.app that transforms extracted con-cepts into navigable concept maps. User eval-uation (n=14) showed that participants experi-enced a 31.5% reduction in perceived cognitiveload when using concept maps, despite spend-ing more time with the visualization (22.6%increase). They also completed comprehensionassessments more efficiently (14.1% faster)with comparable accuracy. This work demon-strates that LLM-based concept mapping cansignificantly reduce cognitive demands whilesupporting non-linear exploration.
Proceedings of the ACL Workshop on Innovative Use of NLP for Building Educational Applications (BEA) / 2025
Anthology | Paper | Poster | BibTeX | GitHub