Improving Abusive Language Detection with online interaction network
作者：
Highlights：
• We consider the rationality of introducing community structural information into Abusive Language Detection task, and propose a pipeline framework to integrate contextual, semantic, and community structural features contained in tweets.
• We build relation-special graphs for tweets as well as comments and design a Relation-Special Graph Neural Network to learn useful information from the graphs.
• We formalize the validity of the proposed framework through experiments and verify that our method can bring greater performance improvement in the case of community density and less training data.
摘要
•We consider the rationality of introducing community structural information into Abusive Language Detection task, and propose a pipeline framework to integrate contextual, semantic, and community structural features contained in tweets.•We build relation-special graphs for tweets as well as comments and design a Relation-Special Graph Neural Network to learn useful information from the graphs.•We formalize the validity of the proposed framework through experiments and verify that our method can bring greater performance improvement in the case of community density and less training data.
论文关键词：Abusive Language Detection,BERT,Social network,Graph neural network,Interaction network
论文评审过程：Received 9 January 2022, Revised 18 June 2022, Accepted 26 June 2022, Available online 8 July 2022, Version of Record 8 July 2022.
论文官网地址：https://doi.org/10.1016/j.ipm.2022.103009
