Seq2RDF: An end-to-end application for deriving Triples from Natural Language Text

  • 2018-07-04 20:13:31
  • Yue Liu, Tongtao Zhang, Zhicheng Liang, Heng Ji, Deborah L. McGuinness
  • 29

Abstract

We present an end-to-end approach that takes unstructured textual input andgenerates structured output compliant with a given vocabulary. Inspired byrecent successes in neural machine translation, we treat the triples within agiven knowledge graph as an independent graph language and propose anencoder-decoder framework with an attention mechanism that leverages knowledgegraph embeddings. Our model learns the mapping from natural language text totriple representation in the form of subject-predicate-object using theselected knowledge graph vocabulary. Experiments on three different data setsshow that we achieve competitive F1-Measures over the baselines using oursimple yet effective approach. A demo video is included.

 

Quick Read (beta)

loading the full paper ...