Defeats GAN: A Simpler Model Outperforms in Knowledge Representation Learning

TL;DR

This paper introduces Trans-DLR, a simple yet effective knowledge representation learning method that outperforms GAN-based approaches by using dynamic learning rate control, a novel negative sampling trick, and efficient evaluation techniques.

Contribution

The paper presents Trans-DLR, a new knowledge embedding model with dynamic learning rate control and relation-aware negative sampling, surpassing existing GAN-based methods.

Findings

Trans-DLR achieves superior performance in knowledge graph tasks.

The negative sampling trick improves model robustness.

Efficient evaluation accelerates link prediction tasks.

Abstract

The goal of knowledge representation learning is to embed entities and relations into a low-dimensional, continuous vector space. How to push a model to its limit and obtain better results is of great significance in knowledge graph's applications. We propose a simple and elegant method, Trans-DLR, whose main idea is dynamic learning rate control during training. Our method achieves remarkable improvement, compared with recent GAN-based method. Moreover, we introduce a new negative sampling trick which corrupts not only entities, but also relations, in different probabilities. We also develop an efficient way, which fully utilizes multiprocessing and parallel computing, to speed up evaluation of the model in link prediction tasks. Experiments show that our method is effective.