Fine-grained Type Prediction of Entities using Knowledge Graph Embeddings

Wikipedia is the largest online encyclopedia, which appears in more than 301 different languages, with the English version containing more than 5.9 million articles. However, using Wikipedia means reading it and searching through pages to find the needed information. On the other hand, DBpedia contains the information of Wikipedia in a structured manner, that is easy to reuse. Knowledge bases such as DBpedia and Wikidata have been recognised as the foundation for diverse applications in the field of data mining, information retrieval and natural language processing. A knowledge base describes real-world objects and the interrelations between them as entities and properties. The entities that share common characteristics are associated with a corresponding type. One of the most important pieces of information in knowledge bases is the type of the entities described. However, it has been observed that type information is often noisy or incomplete. In general, there is a need for well-defined type information for the entities of a knowledge base. In this thesis, the task of fine-grained entity typing of entities of a knowledge base, more specifically - DBpedia, is addressed. ... mehrThere are a lot of entities in DBpedia that are not assigned to a fine-grained type information, rather assigned to either coarse-grained type information or to rdf:type owl:Thing. Fine-grained entity typing aims at assigning more specific types, which are more informative than the coarse-grained ones. This thesis explores and evaluates different approaches for type prediction of entities in DBpedia - the unsupervised approach vector similarity using knowledge graph embeddings, as well as the supervised one - CNN classification. Knowledge graph embeddings from the pre-trained RDF2Vec model are used.