Conservative Knowledge Graph Completion Using Semantically Enriched Information

Anonymous

Abstract: In this paper, we present a novel conservative completion approach for Knowledge Graphs (KGs), designed to address the shortcomings of current knowledge completion methods, particularly their failure to guarantee the accuracy of completion results. Our method uniquely utilizes semantically enriched information inherent in KGs to construct a reasoner based on description logic. By integrating this reasoner with Link Prediction (LP) models, we ensure the correctness of the knowledge completion. Experimental findings show that a substantial proportion of predictions from diverse LP models can undergo conservative completion. Additionally, the volume of conservatively completable results escalates with the increase in semantically enriched information in the KGs.

Paper Type: long

Research Area: Semantics: Sentence-level Semantics, Textual Inference and Other areas

Contribution Types: Model analysis & interpretability

Languages Studied: English