Knowledge graph embedding with self adaptive double-limited loss

作者：

Highlights：

• A new loss function, named ADL, for Knowledge graph embedding is proposed.

• Incorporating ADL into the currently popular embedding models, five extensions, i.e., TransE-ADL, TransH-ADL, TransD-ADL, TorusE-ADL, and ComplEx-ADL, are developed.

• The developed models have significantly improved the performance of link prediction in comparison with the baseline model.

摘要

•A new loss function, named ADL, for Knowledge graph embedding is proposed.•Incorporating ADL into the currently popular embedding models, five extensions, i.e., TransE-ADL, TransH-ADL, TransD-ADL, TorusE-ADL, and ComplEx-ADL, are developed.•The developed models have significantly improved the performance of link prediction in comparison with the baseline model.

论文关键词：Knowledge graph,Embedding,Loss function,Negative sampling,Hard pair

论文评审过程：Received 19 January 2022, Revised 19 June 2022, Accepted 20 June 2022, Available online 25 June 2022, Version of Record 2 July 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.109310