Sentiment Pulse - Multi-Class Sentiment Analysis

🚀 Overview

🧠 Sequential Intelligence

• BiLSTM (Default): A Bidirectional Long Short-Term Memory network that reads text from both ends to preserve context.
• Transformer Encoder: Uses self-attention mechanisms to weigh the importance of different words in a sentence, regardless of their distance.

📁 Key Components

• model.py: BiLSTM & Transformer Encoder implementations.
• train.py: Multi-class training pipeline with CrossEntropyLoss and per-class metrics.
• app.py: Flask inference server with dynamic model selection.
• static/ & templates/: Premium UI with probability distribution visualization.

🛠 Setup & Usage

1. Install Dependencies:
   pip install -r requirements.txt
2. Train the Model:
   bash

   1# Train default BiLSTM 2python train.py --model bilstm 3 4# Or train Transformer 5python train.py --model transformer
3. Launch the Web App:
   python app.py

📊 Performance & Explainability

• Target Accuracy: 80-90%
• Loss Function: CrossEntropyLoss
• Explainability: The system uses a Self-Attention layer. Each word in the input is assigned a weight representing its importance to the final prediction. These weights are used in the web UI to highlight influential words.
• REST API: Provides a /api/predict endpoint for JSON-based sentiment and explainability data.

🌐 Modern UI

• Shimmering Titles: Dynamic text animations for a premium feel.
• Distribution Bars: Real-time breakdown of model confidence across all categories.
• Floating Blobs: Interactive background elements.