Bert is a Transformer Bidirectional Encoder based Architecture trained on MLM(Mask Language Modeling) objective. bert-base-uncased finetuned on the cyrilzhang/financial_phrasebank_split dataset using HuggingFace Trainer with below training parameters.

 learning rate 2e-5, 
 batch size 8,
 num_train_epochs=6,

Model Performance

Epoch	Training Loss	Validation Loss	Accuracy	F1
6	0.034100	0.954745	0.853608	0.854358

How to Use the Model

python

from transformers import pipeline
nlp = pipeline(task='text-classification', 
               model='nickwong64/bert-base-uncased-finance-sentiment')
p1 = "HK stocks open lower after Fed rate comments"
p2 = "US stocks end lower on earnings worries"
p3 = "Muted Fed, AI hopes send Wall Street higher"
print(nlp(p1))
print(nlp(p2))
print(nlp(p3))
"""
output:
[{'label': 'negative', 'score': 0.9991507530212402}]
[{'label': 'negative', 'score': 0.9997240900993347}]
[{'label': 'neutral', 'score': 0.9834381937980652}]
"""

Dataset

cyrilzhang/financial_phrasebank_split

Labels

{0: 'negative', 1: 'neutral', 2: 'positive'}

Evaluation

{'test_loss': 0.9547446370124817,
 'test_accuracy': 0.8536082474226804,
 'test_f1': 0.8543579048224414,
 'test_runtime': 4.9865,
 'test_samples_per_second': 97.263,
 'test_steps_per_second': 12.233}