From unstructured to linked data: Entity extraction and disambiguation by collective similarity maximization

In this paper, we describe a pipeline of methods for identifying and resolving entities from unstructured data using a semi-structured background knowledge database. For this purpose, we employ named entity extraction, co-reference resolution and investigate performance of disambiguation using collective maximization of inter-entity similarity, compared to using only pair-wise disambiguation. We explore possibilities of using DBpedia and Yago as background knowledge databases with the goal of annotating unstructured text documents with global entity references.