You can deploy this model and start earning money today!

TahianaAndriambahoaka/
sentiment-analysis-food-reviews

Name: sentiment-analysis-food-reviews
Availability: InStock

safetensors

roberta

Views

No views yet

yaml

---
language: en
license: mit
base_model: cardiffnlp/twitter-roberta-base-sentiment-latest
tags:
- sentiment-analysis
- food-reviews
- roberta
- text-classification
- fine-tuned
datasets:
- food-com-recipes-and-user-interactions
pipeline_tag: text-classification
widget:
- text: "This recipe was absolutely delicious! Will make again."
  example_title: "Positive review"
- text: "Too salty and instructions were confusing."
  example_title: "Negative review"
- text: "It was okay, nothing special."
  example_title: "Neutral review"
---

RoBERTa Sentiment Analysis - Food Reviews

Model Description

This model is a fine-tuned version of cardiffnlp/twitter-roberta-base-sentiment-latest specifically adapted for analyzing sentiments in food recipe reviews.

The base model (RoBERTa) was originally trained on a large corpus of Twitter data and has been fine-tuned on the Food.com recipes dataset to better understand the specific language and expressions used in culinary reviews.

Intended Use

This model is designed to classify food recipe reviews into three sentiment categories:

negative: Negative sentiment
neutral: Neutral sentiment
positive: Positive sentiment

Usage

Using the pipeline API (recommended)

python

from transformers import pipeline

# Load the model
analyzer = pipeline(
    "sentiment-analysis",
    model="TahianaAndriambahoaka/sentiment-analysis-food-reviews"
)

# Analyze a review
result = analyzer("This recipe is amazing!")
print(result)
# Output: [{'label': 'positive', 'score': 0.9876}]

### Batch processing multiple reviews

```python
reviews = [
    "Amazing! Will make again.",
    "Too salty, not recommended.",
    "It was okay, nothing special."
]

results = analyzer(reviews)
for review, result in zip(reviews, results):
    print(f"{review}: {result['label']} ({result['score']:.2%})")

Using the model directly

python

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "TahianaAndriambahoaka/sentiment-analysis-food-reviews"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

text = "This recipe is amazing!"
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=256)

with torch.no_grad():
    outputs = model(**inputs)
    predictions = torch.nn.functional.softmax(outputs.logits, dim=-1)
    predicted_class = torch.argmax(predictions, dim=-1).item()

label_map = {0: "negative", 1: "neutral", 2: "positive"}
print(f"Sentiment: {label_map[predicted_class]}")

Labels

The model outputs one of three labels:

negative: Reviews expressing dissatisfaction, criticism, or negative experiences
neutral: Reviews that are balanced, informative, or lack strong sentiment
positive: Reviews expressing satisfaction, praise, or positive experiences

Training Data

The model was fine-tuned on the Food.com Recipes and User Interactions dataset, which contains:

User reviews of food recipes
Star ratings (1-5 stars)
Sentiment labels derived from ratings:
- 1-2 stars → Negative
- 3 stars → Neutral
- 4-5 stars → Positive

Training Procedure

Base Model

Base architecture: RoBERTa (Robustly Optimized BERT Approach)
Pre-trained model: cardiffnlp/twitter-roberta-base-sentiment-latest
Fine-tuning task: Sentiment classification (3 classes)

Training Hyperparameters

Epochs: 5
Batch size: 16 (train), 32 (eval)
Learning rate: Default AdamW with warmup
Max sequence length: 256 tokens
Optimizer: AdamW with weight decay 0.01
Warmup steps: 500

Preprocessing

Reviews are tokenized using the RoBERTa tokenizer
Maximum length: 256 tokens
Padding and truncation applied

Performance

The model achieves strong performance on food review sentiment classification. Evaluation metrics on the test set:

Accuracy: High accuracy in classifying sentiments
F1-Score: Balanced performance across all three classes

Limitations

The model is specifically trained on food recipe reviews and may not generalize well to other domains
Reviews in languages other than English may not be classified accurately
Very short reviews (1-2 words) may have lower confidence scores
The model may struggle with highly sarcastic or ironic reviews

Citation

If you use this model, please cite the original dataset:

@article{food-com-dataset,
  title={Food.com Recipes and User Interactions},
  author={Shuyang Li},
  year={2019},
  url={https://www.kaggle.com/shuyangli94/food-com-recipes-and-user-interactions}
}

Model Card Authors

TahianaAndriambahoaka

Contact

For questions or feedback, please open an issue on the model repository.