A joint learning approach with knowledge injection for zero-shot cross-lingual hate speech detection
作者：
Highlights：
• We propose a joint-learning architecture for cross-lingual hate speech detection.
• Our zero-shot approach transfers knowledge between different languages.
• We leverage one resource-rich language to inform models for lower-resource ones.
• We experiment on six lower-resource target languages.
• We experiment with three different multilingual language representation models.
• We investigate the impact of an external resource for knowledge transfer.
• We investigate the creative use of language conveying a derogatory meaning.
摘要
•We propose a joint-learning architecture for cross-lingual hate speech detection.•Our zero-shot approach transfers knowledge between different languages.•We leverage one resource-rich language to inform models for lower-resource ones.•We experiment on six lower-resource target languages.•We experiment with three different multilingual language representation models.•We investigate the impact of an external resource for knowledge transfer.•We investigate the creative use of language conveying a derogatory meaning.
论文关键词：Hate speech detection,Cross-lingual classification,Social media,Transfer learning,Zero-shot learning
论文评审过程：Received 1 November 2020, Revised 23 December 2020, Accepted 7 February 2021, Available online 1 March 2021, Version of Record 1 March 2021.
论文官网地址：https://doi.org/10.1016/j.ipm.2021.102544
