EO4GEO BoK annotation of GI resources

Abstract

The Earth Observation for Geospatial Information (EO4GEO) Body of Knowledge (BoK) serves as a foundational framework encompassing essential geospatial concepts necessary for leveraging Geographic Information and Earth Observation data effectively. Based on the BoK a set of tools was developed at the University Jaume I, including the BoK Annotation Tool (BAT), which facilitates the annotation of any PDF document with EO4GEO BoK concepts, streamlining the process of knowledge association. These annotations are added manually through an easy-to-use “what you see is what you get” (WYSIWYG), which presents significant challenges, including time consumption and the need for domain expertise. To address this challenge, this master thesis studies the use of Natural Language Processing (NLP) techniques to automate the BoK annotation process. Concretely twelve NLP-based tools were applied, utilizing three key phrase extraction algorithms (YAKE, PatternRank, KeyBert) and four semantic similarity measures (Cosine, Jaro-Wrinkler, Latent semantic similarity, Word2Vec), in order to (semi)-automatically generate BoK annotation. To assess the performance, a comparative evaluation was carried out using various annotation approaches (i.e. using top-level concepts, leaf concepts and all concepts) and evaluation methods (i.e. direct matching, parent-child matching, ranking). Results revealed that YAKE_JaroW emerges as a standout performer (F1-score 28.28%), particularly in the parent-child evaluation method. This research helps to annotate existing resources with EO4GEO BoK concepts easier and more efficiently, helping to share knowledge more effectively in geospatial fields. It also emphasizes the importance of having annotation tools designed specifically for the EO4GEO BoK, which fills a crucial gap in geospatial knowledge management.