Distant supervision for treatment relation extraction by leveraging MeSH subheadings

作者：

Highlights：

• MeSH subheadings are leveraged to generate examples from PubMed abstracts for distantly supervised treatment relation extraction.

• MeSH Subheadings based Distant Supervision (MSDS) resulted in 17% improvement in PR-AUC over traditional distant supervision.

• Examples from MSDS can be used to augment crowd-sourced examples for additional performance gains.

• Incorporating a noise-resistant loss function improves performance of models trained on examples generated by MSDS.

摘要

•MeSH subheadings are leveraged to generate examples from PubMed abstracts for distantly supervised treatment relation extraction.•MeSH Subheadings based Distant Supervision (MSDS) resulted in 17% improvement in PR-AUC over traditional distant supervision.•Examples from MSDS can be used to augment crowd-sourced examples for additional performance gains.•Incorporating a noise-resistant loss function improves performance of models trained on examples generated by MSDS.

论文关键词：Relation extraction,Medical treatment relation,Distant supervision,MeSH subheadings

论文评审过程：Received 16 August 2018, Revised 4 June 2019, Accepted 5 June 2019, Available online 7 June 2019, Version of Record 4 July 2019.

论文官网地址：https://doi.org/10.1016/j.artmed.2019.06.002