Movie Review Sentiment Analysis

This project is a sentiment analysis model for movie reviews using DistillBERT. It aims to classify movie reviews as positive or negative based on the sentiment expressed in the text.

Model Information

Data

1. Source

The dataset is from IMDB (Internet Movie Database), a comprehensive online database of films, TV shows, and reviews. IMDB is widely recognized for its extensive movie ratings and critiques from both professional critics and a large community of movie enthusiasts.

2. Data Pre-processing

• Handling missing data: Remove any incomplete entries to maintain data integrity.
• Data normalization:
  • Convert labels: 1 for Positive and 0 for Negative to streamline training.
  • Remove HTML tags: Using BeautifulSoup to clean unnecessary tags from text.
  • Remove emojis: Emojis add sentiment value but introduce inconsistencies in text processing. Regular expressions (RE) are used to filter them out.
• Splitting data:
  • 80%: Training data
  • 16%: Validation data
  • 4%: Test data

Model Performance

1. Training Phase

The model is trained with EarlyStoppingCallback = 3 and evaluated every 500 steps. Below are the key results:

Step	Training Loss	Validation Loss	Accuracy	F1-score
500	0.2996	0.2781	0.8958	0.8958
1000	0.2623	0.2763	0.8978	0.8977
1500	0.2519	0.2579	0.9050	0.9049
2000	0.2318	0.2506	0.9045	0.9044
2500	0.2382	0.2615	0.9031	0.9030
3000	0.1658	0.2886	0.9095	0.9095
3500	0.1700	0.3339	0.9069	0.9067
4000	0.1713	0.3051	0.9116	0.9116
4500	0.1691	0.2852	0.9031	0.9030
5000	0.1825	0.2443	0.9148	0.9147
5500	0.1069	0.3562	0.9148	0.9147
6000	0.1058	0.3747	0.9140	0.9140
6500	0.1076	0.3751	0.9071	0.9070

Running the Model

1. Running on Kaggle

You can use Kaggle's free GPU resources to train and test the model:

• Visit Kaggle: Movie Review Sentiment Analysis

• Click "Copy & Edit" on the upper right corner.

• In Session Options, enable GPU (recommended: T4 x2 for better performance).

  

• Click "Run All" and wait (~30 minutes) for the training phase to complete.

2. Sentiment Analysis (Testing the Model)

Modify the last code cell with your own text input to predict sentiment. Example:

So hot today =_= don't like it and I hate my new timetable, having such a bad week

Run the cell (Ctrl + Enter) to get the sentiment prediction:

Negative

3. Running on Local VS Code

Installation

1. Clone the repository:

   git clone https://github.com/HungDongne/Movie-review-sentiment-analysis-by-using-DistillBERT.git

2. Set up a virtual environment (recommended to prevent conflicts):

   python -m venv .env

3. Activate the environment:

   • On Windows:

     .env\Scripts\activate

   • On macOS/Linux:

     source .env/bin/activate

4. Install dependencies:

   pip install -r requirements.txt

5. Deactivate the environment when done:

   deactivate

Usage

Run the script and modify the last cell to test with your own sentences.

License

This project is licensed under the MIT License. Feel free to use and modify it for your needs!