Fast Linear Model for Knowledge Graph Embeddings

TL;DR

This paper demonstrates that a simple Bag-of-Words approach using fastText can achieve state-of-the-art knowledge graph embeddings efficiently by modeling entity-relation co-occurrences as supervised classification tasks.

Contribution

It introduces a fast, simple baseline method for knowledge graph embeddings that rivals complex models in performance and training time.

Findings

Achieves state-of-the-art results in knowledge graph completion

Training time is reduced to a few minutes

Modeling co-occurrences effectively captures relational information

Abstract

This paper shows that a simple baseline based on a Bag-of-Words (BoW) representation learns surprisingly good knowledge graph embeddings. By casting knowledge base completion and question answering as supervised classification problems, we observe that modeling co-occurences of entities and relations leads to state-of-the-art performance with a training time of a few minutes using the open sourced library fastText.