Cost-effective CNNs-based prototypical networks for few-shot relation classification across domains

作者：

Highlights：

• We enhance prototypical networks for cross-domain few-shot relation classification.

• A multi-channel encoder is introduced inspired by the human recognition process.

• Tree-based attention and selecting strategies are proposed to tackle domain variation.

• The proposed models are proven faster and more cost-effective than baseline models.

摘要

•We enhance prototypical networks for cross-domain few-shot relation classification.•A multi-channel encoder is introduced inspired by the human recognition process.•Tree-based attention and selecting strategies are proposed to tackle domain variation.•The proposed models are proven faster and more cost-effective than baseline models.

论文关键词：Relation classification,Few-shot learning,Domain adaptation,Prototypical network

论文评审过程：Received 4 November 2021, Revised 13 July 2022, Accepted 13 July 2022, Available online 28 July 2022, Version of Record 9 August 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.109470