Table 3 Testing results based on the accuracy, precision, recall, and F1 on non-HS and HS classes
From: An approach to automatic classification of hate speech in sports domain on social media
Test dataset | Epochs | Precision non-HS / HS | Recall non-HS / HS | F1 non-HS/HS | Weighted avg Acc |
|---|---|---|---|---|---|
Training dataset: YouTube entertainment channels—dataset 1 in Table 1 | |||||
YouTube entertainment channels (2) | 5 | 0.29/0.68 | 0.80/0.18 | 0.43/0.28 | 0.36 |
News portals – sports news (5) | 5 | 0.64/0.40 | 0.62/0.42 | 0.63/0.41 | 0.54 |
YouTube sports channels (3) | 5 | 0.04/0.97 | 0.91/0.13 | 0.08/0.23 | 0.23 |
Training dataset: News portals sports comments—dataset 4 in Table 1 | |||||
YouTube entertainment channels (2) | 5 | 0.29/0.68 | 0.78/0.20 | 0.42/0.31 | 0.34 |
News portals – sports news (5) | 5 | 0.60/0.23 | 0.87/0.07 | 0.71/0.17 | 0.48 |
YouTube sports channels (3) | 5 | 0.04/0.96 | 0.91/0.10 | 0.08/0.18 | 0.18 |