Exploration of TransE in a Distributed Environment

Meiyan Lu, Lejian Liao, Feng Zhang, Dandan Song

Abstract: Knowledge graph is popular in knowledge mining fields. TransE uses the structure information of triples (→eh + →er ≈ →et) to embed knowledge graphs into a continuous vector space, which is a very important component in knowledge representations. However, current TransE models are only implemented on single-node machines. With the explosive growth of data volumes, single-node TransE cannot meet the demand for data processing of large knowledge graphs, so a distributed TransE is urgently needed. In this poster, we propose a distributed TransE written in MPI, which can run on HPC clusters. In our experiments, our distributed TransE exhibits high-performance speedup and accuracy.