Learning to classify relations between entities from noisy data - A meta instance reweighting approach
作者：
Highlights：
• Meta-learning could reweight noisy data for robust neural model training.
• Manually annotated reference instances guide meta instance reweighting.
• The inadequacy of expression coverage in reference data harms meta-reweighting.
• Diverse elite instances from noisy data increase coverage of reference data.
• Augmented reference data enhance meta instance reweighting.
摘要
•Meta-learning could reweight noisy data for robust neural model training.•Manually annotated reference instances guide meta instance reweighting.•The inadequacy of expression coverage in reference data harms meta-reweighting.•Diverse elite instances from noisy data increase coverage of reference data.•Augmented reference data enhance meta instance reweighting.
论文关键词：Relation classification,Meta-learning,Instance weighting,Noisy label,Reference data
论文评审过程：Received 18 August 2021, Revised 18 January 2022, Accepted 28 March 2022, Available online 22 April 2022, Version of Record 6 May 2022.
论文官网地址：https://doi.org/10.1016/j.eswa.2022.117113
