crisis_sentiment_roberta

This model is a fine-tuned version of finiteautomata/bertweet-base-sentiment-analysis on an unknown dataset. It achieves the following results on the testing set:

• Accuracy: 0.83
• Macro accuracy: 0.76
• Weighted accuracy: 0.83

Model description

0. Negative
1. Positive
2. Neutral

Sentiment classification using 9,300 tweets of the Flint Water Crisis

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.4781	1.0	349	0.4452	0.8366
0.2074	2.0	698	0.5010	0.8237
0.047	3.0	1047	0.5772	0.8199
0.0114	4.0	1396	0.7793	0.8226
0.007	5.0	1745	0.8584	0.8188
0.0144	6.0	2094	0.9517	0.8070
0.0017	7.0	2443	1.0054	0.8231
0.0013	8.0	2792	1.1297	0.8172
0.0008	9.0	3141	1.1622	0.8263
0.001	10.0	3490	1.2313	0.8204
0.0006	11.0	3839	1.2360	0.8220
0.0007	12.0	4188	1.2687	0.8161
0.0004	13.0	4537	1.2940	0.8204
0.0451	14.0	4886	1.3163	0.8194
0.0004	15.0	5235	1.2991	0.8242

Framework versions

• Transformers 4.23.0.dev0
• Pytorch 1.13.0.dev20220917+cu117
• Datasets 2.4.0
• Tokenizers 0.12.1