Abuse Detection on Cyberbullying Dataset 3 classes (tweet-level)
96AUCDL-Text & Metadata (Interleaved)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DL-Text & Metadata (Interleaved)Features=Text & Metadata, Training Strategy=Interleaved Training2018.02 | 96 | 92 | 93 | 92 | 93 | |
| DL-Text & Metadata (Tran. Lear.)Features=Text & Metadata, Training Strategy=Transfer Learning2018.02 | 95 | 90 | 92 | 90 | 90 | |
| DL-Text & Metadata (Tran. Lear. FT)Features=Text & Metadata, Training Strategy=Transfer Learning with Fine-Tuning2018.02 | 95 | 90 | 91 | 90 | 91 | |
| DL-Text & Metadata (Naive Train.)Features=Text & Metadata, Training Strategy=Naive Training2018.02 | 94 | 89 | 90 | 90 | 90 | |
| DL-Metadata onlyFeatures=Metadata only2018.02 | 93 | 88 | 91 | 88 | 89 | |
| DL-Text onlyFeatures=Text only2018.02 | 92 | 89 | 91 | 89 | 89 | |
| Chatzakou et al. 2017Level=User-level2018.02 | 91 | 91 | 90 | 92 | 91 | |
| Baseline Naive Bayes2018.02 | 73 | 88 | 88 | 88 | 88 |