Large-scale Semantic Annotation for Improved Content Discovery in a Digital Library: A Case Study on Érudit

TitleLarge-scale Semantic Annotation for Improved Content Discovery in a Digital Library: A Case Study on Érudit
Publication TypeReport
Year of Publication2022
AuthorsGotti, F., P. Langlais, and V. Letard
Date Published08/2022
Keywordsdigital libraries, exploratory search, Natural language processing, Open information extraction, semantic annotation, semantic web
AbstractIn this work, we describe through a case study how two natural language processing techniques, end-to-end entity linking and Open Information Extraction (OIE), can be combined to perform semantic annotation of scholarly documents, with a view to improving the discoverability of their content. Our case study is carried out via the implementation of this hybrid approach within a fully realized prototype, Allium, whose primary goal is to facilitate content discovery and navigation by a human user within the Érudit digital library. Érudit comprises over 150 scholarly journals and 38 cultural publications in social sciences and humanities, from Québec and Canada. We start by showing that entity linking allows for a solid foothold in Linked Open Data (LOD), even if it lacks in recall, especially regarding relations between entities. We then show the potential of OIE for content discovery, in part because it can complement the information gleaned from LOD. We propose and implement methods of integrating these elements to a full-fledged publication platform. Finally, we perform a system-oriented assessment and a user-oriented (human) evaluation confirming that LOD and OIE annotations are compatible and complementary for content discovery.