Joint Learning of Entity and Type Embeddings for Analogical Reasoning with Entities

Two situations can be considered analogous if they share a common pattern. Analogical reasoning is the task of finding analogies and inferencing missing terms in them. Since natural language is ambiguous, as the same word can refer to different entities, the use of disambiguated entities from Knowledge Graphs for analogical reasoning might bring to better results. Also, entities have types, i.e. classes, in an ontology, from which they inherit characteristics and properties. In this work we focus on a method to represent entities and their types in a joint vector space to do analogical reasoning. We experiment our representations on a dataset that contains analogies on entities and we show that extending the entity representations with information coming from the types improves analogical reasoning results.