You can deploy this model and start earning money today!

xShubh007/
sentiment_analysis

Name: sentiment_analysis
Availability: InStock

Views

No views yet

🧠 SentimentAI — Full Python Project v2

A production-ready Sentiment Analysis pipeline with 5 ML models, BERT transformer, hyperparameter tuning, and a Flask web dashboard.

📁 Project Structure

sentiment_analysis/
├── main.py                    ← Entry point (run this!)
├── app.py                     ← Flask Web UI
│
├── data_preprocessing.py      ← Text cleaning + TF-IDF
├── model_training.py          ← 5 ML models + cross-validation
├── evaluation.py              ← F1, Precision, Recall, Confusion Matrix
├── inference.py               ← Predict new text (CLI)
├── bert_model.py              ← BERT zero-shot + fine-tuning
├── hyperparameter_tuning.py   ← GridSearchCV + Optuna
│
├── requirements.txt
└── outputs/                   ← Auto-created
    ├── best_model.pkl
    ├── vectorizer.pkl
    ├── tuned_best_model.pkl
    ├── bert_finetuned/
    ├── confusion_matrix_*.png
    ├── model_comparison.png
    ├── grid_search_results.png
    └── optuna_history.png

⚙️ Installation

pip install -r requirements.txt

🚀 Usage

Command	What it does
`python main.py`	Train all 5 ML models on sample data
`python main.py --csv data.csv`	Train on your own CSV
`python main.py --tune`	Train + GridSearch + Optuna tuning
`python main.py --bert`	Train + BERT zero-shot predictions
`python main.py --bert --finetune`	Train + fine-tune BERT on your data
`python main.py --all`	Run the entire pipeline
`python main.py --predict`	Interactive CLI predictor
`python main.py --web`	Launch Flask web dashboard
`python app.py`	Launch web UI directly

🌐 Web UI (Flask)

bash

# Step 1 — train first
python main.py

# Step 2 — launch UI
python app.py
# Open: http://localhost:5000

Features:

Single text analysis with confidence bar charts
Batch analysis (paste multiple lines, get summary counts)
Switch between ML Ensemble and BERT (if loaded)
Inference latency display
Dark theme dashboard

🤖 BERT Transformer

Zero-shot (no training needed, downloads ~250MB model):

python main.py --bert

Fine-tune on your labelled data (trains distilbert-base-uncased):

python main.py --bert --finetune --csv your_data.csv

🔬 Hyperparameter Tuning

python main.py --tune

Runs:

GridSearchCV — exhaustive search over LR, SVM, Random Forest params
Optuna (Bayesian/TPE) — smart search over continuous param spaces

Saves:

outputs/grid_search_results.png
outputs/optuna_history.png
outputs/tuned_best_model.pkl

📊 Models & Metrics

ML Models trained:

Logistic Regression
Naive Bayes
Linear SVM
Random Forest
Gradient Boosting

Metrics computed per model:

Metric	Description
Accuracy	% correct overall
Precision (weighted)	Weighted exactness
Recall (weighted)	Weighted coverage
F1 Weighted	Main metric
F1 Macro	Unweighted avg F1
F1 Micro	Global F1
F1 Per Class	Per sentiment class
CV F1	5-fold cross-val F1
Confusion Matrix	Visual breakdown

📄 CSV Format

csv

text,label
"Great product!",positive
"Terrible service.",negative
"It was okay.",neutral

Labels must be exactly: positive, negative, neutral

xShubh007/sentiment_analysis