README.md

---
license: mit
datasets:
- cornell-movie-review-data/rotten_tomatoes
language:
- en
base_model:
- distilbert/distilbert-base-uncased
pipeline_tag: text-classification
---


# Rotten Tomatoes Sentiment DistilBERT

This model is a fine-tuned version of `distilbert-base-uncased` for binary sentiment classification on movie review snippets from the Rotten Tomatoes dataset. The model predicts whether a given movie review sentence expresses a **positive** or **negative** sentiment.

## Model Details

### Base Model

The base model is `distilbert-base-uncased`, a smaller and faster distilled version of BERT. It is designed to retain much of BERT’s language understanding ability while being more lightweight.

### Fine-Tuned Task

This model was fine-tuned for **text classification**, specifically **binary sentiment analysis**.

The two output labels are:

| Label ID | Label |
| -------- | -------- |
| 0 | NEGATIVE |
| 1 | POSITIVE |

## Dataset

This model was fine-tuned using the `cornell-movie-review-data/rotten_tomatoes` dataset from Hugging Face Datasets.

The dataset contains short movie review sentences labeled as either positive or negative.

Example inputs:

```text
"a smart, funny and touching film"
```

```text
"the story is dull and lifeless"
```

## Intended Use

This model can be used for simple sentiment classification of short English movie-review-style text.

Example use cases include:

* Classifying movie reviews as positive or negative
* Learning and demonstrating Hugging Face Transformers fine-tuning
* Experimenting with the `Trainer` API
* Building a basic sentiment analysis demo

## Limitations

This model was trained on movie review data, so it may perform best on text that resembles short movie reviews.

It may not perform well on:

* Long documents
* Non-English text
* Sarcasm or highly ambiguous sentiment
* Domain-specific sentiment outside movie reviews
* Reviews with mixed opinions

For example, a sentence like:

```text
"The acting was excellent, but the story was slow and forgettable."
```

may be difficult because it contains both positive and negative sentiment.

## Training Procedure

The model was fine-tuned using the Hugging Face Transformers `Trainer` API.

### Training Configuration

```python
TrainingArguments(
output_dir="./rt-sentiment-distilbert",
learning_rate=2e-5,
per_device_train_batch_size=16,
per_device_eval_batch_size=16,
num_train_epochs=3,
weight_decay=0.01,
eval_strategy="epoch",
save_strategy="epoch",
load_best_model_at_end=True,
metric_for_best_model="accuracy",
report_to="none",
)
```

> FULL TRAINING AND FINETUNING CODE CAN BE ACCESSED IN THIS COLAB NODE BOOK: https://colab.research.google.com/drive/12cwKhvX9LAq39LGk-WbxzchYPla_yrA4?usp=sharing

## Usage

You can use this model with the Hugging Face `pipeline` API:

```python
from transformers import pipeline

classifier = pipeline(
"text-classification",
model="jimjunior/rt-sentiment-distilbert"
)

result = classifier("The movie was touching, funny, and beautifully acted.")
print(result)
```

Example output:

```python
[{"label": "POSITIVE", "score": 0.98}]
```

You can also classify negative text:

```python
classifier("The story was boring and the acting was terrible.")
```

Example output:

```python
[{"label": "NEGATIVE", "score": 0.97}]
```

## Model Output

The model returns one of two labels:

* `NEGATIVE`
* `POSITIVE`

Each prediction also includes a confidence score.