LOD Construction Through Supervised Web Relation Extraction and Crowd Validation

摘要

Free, unstructured text is the dominant format in which information is stored and published. To interpret such vast amount of data one must employ a programmatic approach. In this paper, we describe a novel approach - a pipeline in which interesting relations are extracted from web portals news texts, stored as RDF triplets, and finally validated by end user via browser extension. In the process, different machine learning algorithms were tested on relation extraction, enhanced with our own set of features and thoroughly evaluated, with excellent precision and recall results compared to models used for semantic knowledge expansion. Building on those results, we implement and describe the component to resolve discovered entities to existing semantic entities from three major online repositories. Finally, we implement and describe the validation process in which RDF triplets are presented to the web portal reader for validation via Chrome extension.