BERT Fine-tuned - Financial Sentiment Analysis Model

This model is a Fine-Tuned version of BERT (bert-base-uncased) It is designed to classify text into positive, neutral, and negative sentiments. The fine-tuning was performed using the Financial Phrase Bank dataset.

Results

It achieves the following results on the evaluation set:

F1 Score: 0.9468
Validation loss: 0.1860

Training Data

The dataset consists of 4840 sentences of the financial phrase bank. The dataset was annotated by 16 people with adequate background knowledge of financial markets.

Training hyperparameters

The following hyperparameters were used during training:

learning rate : 2e-5
train_batch_size : 32
eval_batch_size: 32
seed: 42
Optimizer : AdamW
num_epochs: 3

Training Results

Epoch	Validation Loss	Accuracy
01	0.1860	0.9468
02	0.1756	0.9424
03	0.1726	0.9432

This model is a part of my thesis: "A Proposal of a Sentiment Analysis Model for Business Intelligence"