Sentiment Analysis API for Product Reviews

🎯 Project Overview

A production-ready sentiment analysis API that classifies product reviews as positive, negative, or neutral.

A live demo can be found here: Demo

P.S. I hate front-end. Don't judge

🚀 Starting the project

Clone the app

You have two options to run the app: clone this Github repo or use Docker.

1. Using Github

mkdir realy-discriptive-name-that-isnt-super-long
cd ./realy-discriptive-name-that-isnt-super-long
# Clone the project
git clone https://github.com/Niazi04/Farsi-Product-Sentiment-API.git .

# Once you've cloned the project, install dependencies:
# Use a virtual environment to avoid version conflicts
pip install -r requirements.txt

# Then run the app:
python app.py

2. Using docker

⚠️ NOTE: The docker image does not include the notebooks for testing or training ML models

If you want to explore the notebooks, clone the repository via GitHub instead.

docker pull nastyhedgeh0g/farsi-product-sentiment-api:FirstStable

docker run -p 5000:5000 nastyhedgeh0g/farsi-product-sentiment-api:FirstStable

To use a different port:

docker run -p 8080:5000 nastyhedgeh0g/farsi-product-sentiment-api:FirstStable

🧪 Test the app

You can simply test the app using curl or any http client

Health Check Endpoint

curl http://localhost:5000/health

Expected response code: 200 OK

Response body should look like:

{
      "dependencies": {
            "svm":"available",
            "vectorizer":"available"
      },
      "model":"not loaded in memory",
      "timestamp":"2026-02-14T19:50:58.760383+00:00"
}

If either 'svm' or 'vectorizer' show "missing", it means they are not found in ./model directory.

Initially 'model' is not loaded in memory. Because I coded it that way. After your first POST request, it will be loaded in memory

🔮 Get Your First Prediction

💡 Tip: Use Postman, Insomnia, or a similar tool for sending Persian text, as terminals may not handle Unicode properly.

This is how your JSON body should look like:

{
      "review": "گوشی واقعا عالی بود"
}

Send a POST request to http://localhost:5000/predict

If all the previous steps were successfull, you should get a response like so:

{
    "polarity": "positive",
    "success": true
}

After your first request, the model remains loaded in memory for faster subsequent responses.

Send a GET request to ./health endpoint and verify this.

Data Source

The dataset used for training this sentiment analysis model is Pars-ABSA, a benchmark corpus for Persian aspect-based sentiment analysis.

📊 Dataset Information

• Name: Pars-ABSA
• Size: 10,002 samples (5,114 positive, 3,061 negative, 1,827 neutral)
• Language: Farsi (Persian)

📥 How to Obtain the Data

Due to licensing and storage constraints, the full dataset is not included in this repository. However for testing purposes, I have included a sample csv file under 'pipeline/data/processed/sample.csv'

The data is already cleaned and processed and is ready to be fed into the already trained model (look at 'model' dir)

To download the dataset

1. Visit the original repository: https://github.com/Titowak/Pars-ABSA
2. Download 'Pars-ABSA_xml.xml'
3. Place the downloaded file in pipline/data/raw/ directory

🎓 Citation

@misc{ataei2019parsabsaaspectbasedsentimentanalysis,
      title={Pars-ABSA: an Aspect-based Sentiment Analysis dataset for Persian}, 
      author={Taha Shangipour Ataei and Kamyar Darvishi and Soroush Javdan and Behrouz Minaei-Bidgoli and Sauleh Eetemadi},
      year={2019},
      eprint={1908.01815},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/1908.01815}
}