Model Details
Bert Model
Model Description
The model used is a bert model with tuned hyper parameters that worked well for the dataset. It works well with the Car_Reviews dataset I have and predicts the Positive, Negative, Neutral Classes perfectly.
Model Sources
Hugging Face BERT Model
Bias, Risks, and Limitations
I think the model is little biased towards positive features and is not perfect while predicting neutral values.
Training Details
I have trained the model with 200 car reviews which are annotated by the annotators and most the reviews contain 50 words.
Training Hyperparameters
learning_rate=2e-5,
per_device_train_batch_size=8,
num_train_epochs=20,
weight_decay=0.01,
save_total_limit=2,
Evaluation
Accuracy is 81 for Test data.
Testing Data
The test data contains 45 reviews which are also clearly annotated.
