Snippet Generation with Reasoning and Embedding Techniques

Abstract

An increasing amount of Linked Open Data is now available on the Web, resulting in the expansion of knowledge graphs as more triples are added to them. To prevent information overload and improve task efficiency, multiple methods exist to summarize a knowledge graph. Entity summarization is one of these methods and involves producing a small subset, a snippet, of the entity description(s), for a given entity or group of entities. The snippet can then be used in tasks, instead of a lengthy entity description. However, entity summarization is limited to instance level entities and cannot produce snippets for properties or classes. These are other components of a knowledge graph that could be of interest, particularly in combination with instance level entities. Therefore, in this thesis, we propose approaches for generating snippets for not only instance level entities, but also properties and classes. We present two approaches: one based on reasoning with RDFS entailment rules and another based on knowledge graph embedding using RDF2Vec. Additionally, we created a benchmark to evaluate the performance of these two approaches. The results, especially for the reasoning-based approach, were promising.