Dataset Link: https://www.kaggle.com/datasets/devalper/haber-tweetlerinin-duygu-analizi-ve-snflandrma
Model | Test Accuracy (Sentiment) | Test Accuracy (Topic) | Test Accuracy (MultiTask-Sentiment) | Test Accuracy (MultiTask-Topic) | Training Loss (Sentiment) | Training Loss (Topic) | Training Loss (MultiTask) | Training Duration (Sentiment) | Training Duration (Topic) | Training Duration (MultiTask) |
---|---|---|---|---|---|---|---|---|---|---|
Tiny Bert | 67% (8 Epoch) | 37% (8 Epoch) | 68% | 30% | 0.61 | 1.85 | 2.47 | 6 seconds | 6 seconds | 8 seconds |
Mini Bert | 69% (8 Epoch) | 54% (8 Epoch) | 70% | 58% | 0.54 | 1.59 | 2.09 | 8 seconds | 8 seconds | 22 seconds |
Small Bert | 88% (8 Epoch) | 64% (8 Epoch) | 79% | 65% | 0.36 | 1.18 | 1.69 | 8 seconds | 16 seconds | 24 seconds |
Medium Bert | 85% (8 Epoch) | 69% (8 Epoch) | 74% | 70% | 0.33 | 1.06 | 1.50 | 24 seconds | 40 seconds | 48 seconds |
Base Bert | 86% (8 Epoch) | 74% (8 Epoch) | 83% | 75% | 0.20 | 0.87 | 1.37 | 48 seconds | 72 seconds | 136 seconds |
Big Bert | 90% (8 Epoch) | 89% (8 Epoch) | 89% | 91% | 0.02 | 0.20 | 0.27 | 48 seconds | 88 seconds | 144 seconds |