Evaluating the effectiveness of embeddings in representing the structure of geospatial ontologies

Nowadays word embeddings are used for many natural language processing (NLP) tasks thanks to their ability of capturing the semantic relations between words. Word embeddings have been mostly used to solve traditional NLP problems, such as question answering, textual entailment and sentiment analysis. This work proposes a new way of thinking about word embeddings that exploits them in order to represent geographical knowledge (e.g., geographical ontologies). We also propose metrics for evaluating the effectiveness of an embedding with respect to the ontological structure on which it is created both in an absolute way and with reference to its application within geolocation algorithms.