Sentiment Analyzer API

This is a sentiment analysis API built with Flask. It uses a logistic regression model trained on a cleaned version of the Sentiment140 dataset to classify Spanish input text as positive, negative, or neutral. The API also provides actionable recommendations based on the sentiment.

---

Features

• Clean and prepare raw tweets using utils/clean_data.py
• Train a logistic regression model with TF-IDF vectorization
• Predict sentiment using a /predict endpoint
• Access model evaluation metrics via a /metrics endpoint
• Modular structure with scripts and models organized in folders
• Ready for local development with pyenv

---

Prerequisites

• Python 3.10.13 (recommended via pyenv)
• pip package manager

If you're using pyenv, the project includes a .python-version file:

pyenv install 3.10.13  # if not already installed
pyenv local 3.10.13

---

Setup

1. Clone the repository:

git clone https://github.com/JhymerMartinez/sentiment_analyzer.git
cd sentiment_analyzer

2. Create and activate a virtual environment (optional but recommended):

python -m venv venv
source venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

---

Prepare the Dataset

1. Download the original Sentiment140 dataset from here.

   • File: training.1600000.processed.noemoticon.csv
   • Place it inside the data/ folder.

2. Run the data cleaning script:

python utils/clean_data.py

This will output a cleaned_dataset.csv file to the data/ directory.

---

Train the Model

Run the training script to train and save the model along with evaluation metrics:

python utils/train_model.py

This will generate:

• models/sentiment_model.pkl → trained model
• metrics/model_metrics.json → evaluation metrics

---

Run the API

Start the Flask server:

python app.py

The API will be available at: http://127.0.0.1:5000

---

API Endpoints

POST /predict

Send text input and get back a sentiment prediction with a recommendation.

Example:

curl -X POST http://127.0.0.1:5000/predict \
  -H "Content-Type: application/json" \
  -d '{"text": "I like this product"}'

Response:

{
  "sentiment": "positive",
  "recommendation": "Keep reinforcing what you're doing well."
}

---

GET /metrics

Fetch the stored model evaluation metrics:

curl http://127.0.0.1:5000/metrics

Example Response:

{
  "accuracy": 0.797,
  "macro avg": {
    "precision": 0.797,
    "recall": 0.797,
    "f1-score": 0.797
  },
  "weighted avg": {
    "precision": 0.797,
    "recall": 0.797,
    "f1-score": 0.797
  }
}

---

Project Structure

sentiment_analyzer/
│
├── app.py
├── requirements.txt
├── .python-version
│
├── utils/
│   ├── clean_data.py
│   └── train_model.py
│
├── data/
│   ├── training.1600000.processed.noemoticon.csv
│   └── cleaned_dataset.csv
│
├── models/
│   └── sentiment_model.pkl
│
└── metrics/
    └── model_metrics.json

---

Notes

• Make sure data/, models/, and metrics/ directories exist before running the scripts.
• The files sentiment_model.pkl, model_metrics.json, and cleaned_dataset.csv are excluded from version control (see .gitignore).

---

License

MIT License