Update README.md

428a4a1 over 3 years ago

4.47 kB

tags:
  - generated_from_trainer
  - financial-tweets-sentiment-analysis
  - sentiment-analysis
  - generated_from_trainer
  - financial
  - stocks
  - sentiment
datasets:
  - zeroshot/twitter-financial-news-sentiment
metrics:
  - accuracy
  - f1
  - precision
  - recall
widget:
  - text: $LOW - Lowe's racks up another positive rating despite recession risk
    example_title: Bullish Sentiment
  - text: $HNHAF $HNHPD $AAPL - Trendforce cuts iPhone estimate after Foxconn delay
    example_title: Bearish Sentiment
  - text: >-
      Coin Toss: Morgan Stanley Raises Tesla Bull Case To $500, Keeps Bear Case
      At $10
    example_title: Neutral Sentiment
model-index:
  - name: finbert-tone-finetuned-fintwitter-classification
    results:
      - task:
          name: Text Classification
          type: text-classification
        dataset:
          name: twitter-financial-news-sentiment
          type: finance
        metrics:
          - type: F1
            name: F1
            value: 0.8838
          - type: accuracy
            name: accuracy
            value: 0.884

finbert-tone-finetuned-fintwitter-classification

This model is a fine-tuned version of yiyanghkust/finbert-tone on Twitter Financial News dataset. It achieves the following results on the evaluation set:

Loss: 1.4078
Accuracy: 0.8840
F1: 0.8838
Precision: 0.8838
Recall: 0.8840

Model description

Model determines the financial sentiment of given tweets. Given the unbalanced distribution of the class labels, the weights were adjusted to pay attention to the less sampled labels which should increase overall performance..

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: 20
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.6385	1.0	597	0.3688	0.8668	0.8693	0.8744	0.8668
0.3044	2.0	1194	0.3994	0.8744	0.8726	0.8739	0.8744
0.1833	3.0	1791	0.6212	0.8781	0.8764	0.8762	0.8781
0.1189	4.0	2388	0.8370	0.8740	0.8743	0.8748	0.8740
0.0759	5.0	2985	0.9107	0.8807	0.8798	0.8796	0.8807
0.0291	6.0	3582	0.9711	0.8836	0.8825	0.8821	0.8836
0.0314	7.0	4179	1.1305	0.8819	0.8811	0.8812	0.8819
0.0217	8.0	4776	1.0190	0.8811	0.8813	0.8816	0.8811
0.0227	9.0	5373	1.1940	0.8844	0.8832	0.8838	0.8844
0.0156	10.0	5970	1.2595	0.8752	0.8768	0.8801	0.8752
0.0135	11.0	6567	1.1931	0.8760	0.8768	0.8780	0.8760
0.009	12.0	7164	1.2154	0.8857	0.8852	0.8848	0.8857
0.0058	13.0	7761	1.3874	0.8748	0.8759	0.8776	0.8748
0.009	14.0	8358	1.4193	0.8740	0.8754	0.8780	0.8740
0.0042	15.0	8955	1.2999	0.8807	0.8800	0.8796	0.8807
0.0028	16.0	9552	1.3428	0.8802	0.8805	0.8817	0.8802
0.0029	17.0	10149	1.3959	0.8807	0.8807	0.8810	0.8807
0.0022	18.0	10746	1.4149	0.8827	0.8823	0.8824	0.8827
0.0037	19.0	11343	1.4078	0.8840	0.8838	0.8838	0.8840
0.001	20.0	11940	1.4236	0.8823	0.8823	0.8825	0.8823

Framework versions

Transformers 4.25.1
Pytorch 1.13.0+cu116
Datasets 2.8.0
Tokenizers 0.13.2