Request Correction

An Open-Resource Knowledge Augmentation for Biomedical Lay Summarization

Automatic summarization aims to generate concise versions of texts while retaining relevant information. Summaries can be either extractive, using direct excerpts, or abstractive, rephrasing content to convey the same meaning. Lay summarization applies abstractive techniques to simplify complex texts, such as scientific literature, for broader audiences, thereby promoting public understanding of specialized knowledge. Prior work shows that knowledge augmentation improves lay summarization. Still, biomedical applications often rely on closed resources like the Unified Medical Language System (UMLS), which require expert curation and are costly to scale. We propose a four-step approach that leverages keyword extraction and DBpedia, an open general domain knowledge base, ideal to bridge the gap between expert and lay knowledge. First, we extract keywords from biomedical texts using YAKE!, a well-established unsupervised method. Second, we query DBpedia using these keywords to retrieve relevant concept entries. Third, we construct a graph of concepts for each document based on cosine similarity between DBpedia entries. Finally, we combine each graph with the original abstract to train a summarization model. Our method achieves competitive performance compared to UMLS-based systems in the eLife dataset (ROUGE-1: 58.44 vs. 60.26, ROUGE-L: 43.45 vs. 45.45), demonstrating that open-resource approaches can provide viable alternatives to licensed knowledge bases while maintaining accessibility for resource-constrained organizations.