Improving data quality from DBpedia for better language generation
Validating DBpedia Triple Sets for Natural Language Generation
Computation and Language
Summary
Many computers use data from a source called DBpedia to create sentences and descriptions about facts. However, some of the pieces of information, called triples, might be wrong or unclear, which makes the generated sentences less accurate. The study shows that by applying special rules to check these triples, it's possible to pick out the correct ones with very high accuracy. They also found ways to get more good triples without including mistakes, making the information better for language generation tasks.
DBpediatriplesNatural Language Generationprecisionrecallentitydata validationproperty definitions
Authors
Mark Andrade, Simon Mille, Anya Belz, Brian Davis
Abstract
We present a study of the quality of individual DBpedia triples from the perspective of Natural Language Generation, and propose and evaluate an approach for collecting entity-specific triple sets that filters out questionable triples while minimizing the loss of correct ones. We show in an evaluation against manually annotated data that with validation rules, it is possible to reach 98% precision in triple selection, and with improvements to a few Property definitions, it is possible to improve recall by 40% without harming precision.