Embedding Entity Pairs through Observed Relations for Knowledge Base Completion

Dirk Weissenborn

Abstract:

In this work we present a novel approach for the utilization of observed
relations between entity pairs in the task of triple argument prediction.
The approach is based on representing observations in a shared, continuous
vector space of structured relations and text. Results on a recent
benchmark dataset demonstrate that the new model is superior to existing
sparse feature models. In combination with state-of-the-art models,
we achieve substantial improvements when observed relations are available.

Conflicts: dfki.de