Ssl Relation Prediction

This repo contains the code accompanying the paper: “Relation Prediction as an Auxiliary Training Objective for Improving Multi-Relational Graph Representations”. We found that incorporating relation prediction into the 1vsAll objective yields a new self-supervised training objective for knowledge base completion (KBC), which results in significant performance improvement (up to 9.9% in Hits@1) by adding as little as 3–10 lines of code. Unleash the true power of your KBC models with the relation prediction objective!