Sentiment analysis of yelp reviews.
Analyzing customer feeback for businesses.
Helping businesses see the information they need in a timely manner to allow them to answer them more succintly.
This model IS NOT a replacement for humans and will still have issues with sarcasm, do not trust it 100% and please double check its findings.

Training and evaluation data

The model was trained and evaluated on the Yelp/yelp-review-full dataset. It uses the full 650,000 rows of training data and evaluates on the included test set.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 7.637168097601297e-06
train_batch_size: 32
eval_batch_size: 64
seed: 42
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1
0.3669	1.0	20313	0.3651	0.8484	0.8436
0.3343	2.0	40626	0.3645	0.8485	0.8495
0.3133	3.0	60939	0.3689	0.8497	0.8481
0.2815	4.0	81252	0.3773	0.8506	0.8492

Framework versions

Transformers 4.48.3
Pytorch 2.6.0+cu124
Datasets 3.4.0
Tokenizers 0.21.0

Evaluation Results

Metric	Value
eval_loss	0.36449944972991943
eval_accuracy	0.84846
eval_f1	0.8494926642060322
eval_runtime	169.6017
eval_samples_per_second	294.808
eval_steps_per_second	4.611
epoch	4.0

Preprocessing

Tokenized using the default distilbert-base-uncased tokenizer

Model Architecture

The model uses the DistilBERT base uncased architecture.

Fine tuning approach

The model was fine tuned using the hugging face trainer api.

Example Usage

python

from transformers import pipeline
model_path = "FinchW/my-yelp-sentiment-model-finetuned"
tokenizer_path = "distilbert/distilbert-base-uncased"
id2label = {
    0: "NEGATIVE",
    1: "NEUTRAL",
    2: "POSITIVE",
}
label2id = {
    "NEGATIVE": 0,
    "NEUTRAL": 1,
    "POSITIVE": 2
}
tokenizer = AutoTokenizer.from_pretrained(tokenizer_path)
model = AutoModelForSequenceClassification.from_pretrained(model_path, num_labels=3, id2label=id2label, label2id=label2id)

pipe = pipeline("sentiment-analysis", model=model, tokenizer=tokenizer, truncation=True) 
#pass single string
print("Single String:")
display(pipe("My food was disgusting!"))
#pass list of strings
print("List of Strings:")
display(pipe(["My food was amazing", "my food was sub-par", "gross"]))