Query Answering over the Polymorphic Web of Data.

Knowledge graphs are a versatile means to gather Semantic Web data and are typically stored and queried with the W3C standards, RDF and SPARQL. Despite the significant progress made in query processing, predicting plausible answers in presence of missing facts remains a challenge. This aspect has been tackled by proposing methods to predict links and solve queries in some reduced fragments of SPARQL. Thus far, I have explored two parallel directions for this thesis. First, I study how to use knowledge graph embedding methods to predict missing facts. In particular, I explore how to combine different knowledge graph embedding methods to improve the quality of the predictions. Second, I study how to connect techniques to query knowledge graphs from these two research areas, namely database query processing, and graph learning. The former techniques provide actual answers of a query, while the latter provide plausible ones. My hypothesis is that I can define a common query interface, based on SPARQL, to provide answers from these polymorphic data sources. To this end, I propose an extension for SPARQL, called polymorphic SPARQL.