Incorporating GAN for Negative Sampling in Knowledge Representation Learning

TL;DR

This paper introduces a GAN-based framework for knowledge representation learning that generates high-quality negative samples, improving the efficiency and accuracy of embedding models in knowledge graphs.

Contribution

It proposes a novel GAN-based negative sampling method that enhances traditional knowledge embedding models by producing more challenging negative samples.

Findings

Outperforms baseline models on triplet classification

Improves link prediction accuracy

Generates high-quality negative samples

Abstract

Knowledge representation learning aims at modeling knowledge graph by encoding entities and relations into a low dimensional space. Most of the traditional works for knowledge embedding need negative sampling to minimize a margin-based ranking loss. However, those works construct negative samples through a random mode, by which the samples are often too trivial to fit the model efficiently. In this paper, we propose a novel knowledge representation learning framework based on Generative Adversarial Networks (GAN). In this GAN-based framework, we take advantage of a generator to obtain high-quality negative samples. Meanwhile, the discriminator in GAN learns the embeddings of the entities and relations in knowledge graph. Thus, we can incorporate the proposed GAN-based framework into various traditional models to improve the ability of knowledge representation learning. Experimental results show that our proposed GAN-based framework outperforms baselines on triplets classification and link prediction tasks.