Sentiment Analysis is a modern Flask web app that classifies user-entered text as Positive, Negative, or Neutral using an Artificial Neural Network (ANN).

Instead of relying on rule-based keyword matching, the model learns contextual patterns from thousands of real reviews — delivering accurate, confidence-scored predictions in real time.

📦 From a simple Jupyter Notebook experiment, this project has been upgraded into a fully interactive web application with:

• ✔ Text Input
• ✔ Real-time Prediction
• ✔ Confidence Score
• ✔ Sentiment Label with Visual Feedback
• ✔ Clean Glassmorphism UI

All in seconds.

Enter Review Text
      ↓
Tokenize + Pad Sequence
      ↓
ANN Model Inference
      ↓
Apply Confidence Threshold
      ↓
Display Sentiment + Confidence Score

• ANN with Embedding + GlobalAveragePooling layers
• 3-class softmax output: Negative / Neutral / Positive
• Tokenizer with OOV token for unseen words

• Predicted sentiment label
• Confidence percentage
• Color-coded visual feedback (green / red / grey)
• Instant JSON response via REST API

• Glassmorphism card design
• Animated gradient background
• Smooth result transitions
• Mobile responsive layout
• Poppins font + custom CSS styling

• Flask REST API (/predict endpoint)
• JSON request/response handling
• Pre-loaded model and tokenizer at startup
• Lightweight and fast inference

Raw Text → Tokenizer → Integer Sequences

Example:

"This movie was great!" → [4, 56, 12, 87]

Sequences → Fixed length (MAX_LEN = 100)

Shorter sequences are zero-padded; longer ones are truncated.

Padded Sequence → Embedding → GlobalAveragePooling → Dense → Softmax

Output:

[negative_prob, neutral_prob, positive_prob]

[negative_prob, neutral_prob, positive_prob]

The model selects the class with the highest confidence score as the final predicted sentiment.

The model currently struggles with accurately detecting Neutral sentiment. The primary reason is the quality and balance of the training dataset — neutral examples are often underrepresented or inconsistently labeled compared to clearly positive or negative reviews.

As a result, borderline texts may be misclassified as positive or negative instead of neutral.

Improving the neutral class requires a cleaner, more diverse, and better-balanced dataset with well-defined neutral examples.

Sentiment-Analysis-using-ANN/
│
├── app.py
├── predict.py
├── requirements.txt
│
├── data/
│   ├── sentiment_data.csv
│   ├── EcoPreprocessed.csv
│   └── train_df.csv
│
├── model/
│   ├── modal.py
│   ├── sentiment_model.keras
│   └── tokenizer.pkl
│
├── notebooks/
│   └── Sentiment-Analysis.ipynb
│
├── templates/
│   └── index.html
│
├── sampleScreenshots/
│
└── README.md

git clone https://github.com/SACHIN-S-2004/Sentiment-Analysis-using-ANN.git
cd Sentiment-Analysis-using-ANN

pip install -r requirements.txt

python app.py

http://127.0.0.1:5000

This project demonstrates:

• ✔ Supervised Deep Learning (ANN)
• ✔ Natural Language Processing (NLP) fundamentals
• ✔ Text tokenization and sequence padding
• ✔ Flask backend development
• ✔ Practical ML model deployment

Give it a star — it helps a lot!