Improving classifier training efficiency for automatic cyberbullying detection with Feature Density
作者：
Highlights：
• Feature Density can be utilized to reduce of the number of required experiments iterations.
• In general, Feature Density seems to have a negative correlation with classifier performance.
• Dependency structures could have potential as features in Neural Networks.
• Dataset complexity cannot be measured with Feature Density alone.
• Linguistic preprocessing can improve classifier performance.
摘要
•Feature Density can be utilized to reduce of the number of required experiments iterations.•In general, Feature Density seems to have a negative correlation with classifier performance.•Dependency structures could have potential as features in Neural Networks.•Dataset complexity cannot be measured with Feature Density alone.•Linguistic preprocessing can improve classifier performance.
论文关键词：Feature density,Dataset complexity,Linguistics,Cyberbullying,Document classification,Preprocessing
论文评审过程：Received 31 October 2020, Revised 6 April 2021, Accepted 28 April 2021, Available online 13 May 2021, Version of Record 13 May 2021.
论文官网地址：https://doi.org/10.1016/j.ipm.2021.102616
