Automated annotation of GIS workflows using knowledge graph embedding (KGE)

Abstract

To solve the “indirect” question-answering problem, the core concepts of spatial information, which distinguish geo-semantics, are used to interpret both questions and workflows (answers). Geographical Information Systems (GIS) processes occurring in these workflows can then be annotated by the core concept data (CCD) types which combine data forms with their semantics. In QuAnGIS the annotations are currently done manually, however, it is a complex and timeconsuming process. In this project, we check ‘How do Knowledge Graph Embeddings (KGE) models help with automated annotation of GIS workflows?’. We test RESCAL and ConvE models and how they behave with geo-specific data. The aim is to check if automatic annotation with use of KGE models is even possible, and if so, what method would be the best. Some of the results were positive in terms of tail prediction and evaluation metrics. Despite that, the automatic annotation is rather challenging, with the current state of the data used in QuAnGIS project.