Semantic Comment Analysis

A production-ready application for semantic analysis of customer feedback using transformer-based NLP models. Provides automated intent classification, sentiment analysis, and semantic similarity scoring.

Overview

This application leverages state-of-the-art natural language processing to analyze customer comments, support tickets, or social media feedback. It classifies intents into predefined categories, scores sentiment polarity, and computes semantic similarity against reference comments.

Key Features

• Intent Classification: Zero-shot classification using BART-MNLI to identify comment types (Bug Report, Feature Request, Question, Praise, Complaint, General Feedback)
• Sentiment Analysis: Embedding-based sentiment scoring with positive, neutral, and negative components
• Similarity Computation: Cosine similarity measurement between comments using sentence embeddings
• Dual Processing Modes: Single text analysis and batch CSV processing
• Interactive UI: Web-based Gradio interface for easy interaction

Architecture

The application follows a modular design pattern with clear separation of concerns:

Semantic/
├── nlp_engine.py      # NLP model management and inference
├── data_handler.py    # CSV processing and batch operations
├── interface.py       # Gradio UI components and callbacks
├── main.py           # Application entry point
├── requirements.txt  # Python dependencies
└── sample_comments.csv  # Example test data

Module Responsibilities

nlp_engine.py

• Model loading and caching (BART-MNLI, MiniLM)
• Intent classification logic
• Sentiment analysis computation
• Semantic similarity calculation

data_handler.py

• CSV file validation and parsing
• Batch processing workflows
• Result aggregation and statistics

interface.py

• Gradio component definitions
• UI layout and styling
• Callback function orchestration

main.py

• Application initialization
• Model pre-loading
• Interface launch configuration

Setup

Requirements

• Python 3.8 or higher
• 4GB RAM minimum (8GB recommended for batch processing)
• Internet connection for initial model downloads

Installation

1. Clone or download the project directory

2. Create a virtual environment:

python -m venv venv

3. Activate the virtual environment:

Windows:

venv\Scripts\activate

Linux/Mac:

source venv/bin/activate

4. Install dependencies:

pip install -r requirements.txt

Note: First-time setup will download transformer models (approximately 1.7GB total). Models are cached locally for subsequent runs.

Usage

Starting the Application

python main.py

The application will:

1. Load ML models into memory
2. Launch a local web server
3. Open automatically at http://127.0.0.1:7860

Single Text Analysis

1. Navigate to the "Single Analysis" tab
2. Enter the comment text to analyze
3. (Optional) Enter a reference "anchor" comment for similarity comparison
4. Adjust the confidence threshold slider (default: 0.5)
5. Click "Analyze" to process

Output:

• Top 3 predicted intents with confidence scores
• Interactive sentiment polarity chart
• Detailed analysis summary with all metrics

Batch CSV Processing

1. Navigate to the "Batch Processing" tab
2. Upload a CSV file (first row must contain column headers)
3. Specify the column name containing comments (e.g., "comment", "text", "feedback")
4. Adjust the confidence threshold if needed
5. Click "Process" to analyze all rows

Output:

• Preview of first 100 processed rows
• Downloadable CSV with appended analysis columns:
  • Sentiment_Score: Compound sentiment (-1 to 1)
  • Predicted_Intent: Primary intent category
  • Confidence: Classification confidence (0 to 1)
  • Positive_Score: Positive sentiment component
  • Neutral_Score: Neutral sentiment component
  • Negative_Score: Negative sentiment component

CSV Format Example

comment,user_id
The app crashes when I try to export data,user_001
Love the new design! Much cleaner interface,user_002
How do I reset my password?,user_003

Technical Details

Models

Intent Classification:

• Model: facebook/bart-large-mnli
• Task: Zero-shot classification
• Size: ~1.6GB
• Inference: CPU-optimized

Semantic Embeddings:

• Model: sentence-transformers/all-MiniLM-L6-v2
• Task: Sentence embeddings
• Size: ~90MB
• Inference: CPU-optimized

Performance

• First run: 2-3 minutes (model downloads)
• Subsequent runs: 10-15 seconds (cached models)
• Single analysis: <1 second per comment
• Batch processing: ~2-3 seconds per comment

Customization

Modifying Intent Labels:

Edit INTENT_LABELS in nlp_engine.py:

INTENT_LABELS = [
    "Your Custom",
    "Intent Categories",
    "Go Here"
]

Changing Sentiment Anchors:

Edit anchor phrases in analyze_sentiment() function in nlp_engine.py.

Development

Project Structure Details

The modular architecture enables:

• Independent testing of components
• Easy swapping of ML models
• UI redesign without touching business logic
• Parallel development across modules

Adding New Features

New Analysis Type:

1. Add inference function to nlp_engine.py
2. Update interface.py to display results
3. Modify data_handler.py if CSV processing needed

UI Modifications:

1. Edit only interface.py
2. Business logic remains untouched

Code Quality

• PEP 8 compliant formatting
• Type hints on all function signatures
• Comprehensive docstrings
• Modular, testable architecture

Troubleshooting

Models fail to load:

• Ensure stable internet connection
• Check available disk space (2GB minimum)
• Try manual download: huggingface-cli download facebook/bart-large-mnli

CSV processing errors:

• Verify first row contains headers
• Check column name spelling
• Ensure CSV is properly formatted (UTF-8 encoding recommended)

Performance issues:

• Reduce batch size for large CSV files
• Close other memory-intensive applications
• Consider GPU acceleration for large-scale deployments

License

This project is provided as-is for educational and commercial use.

Support

For issues or questions, verify:

1. All dependencies installed correctly
2. Python version 3.8+
3. Models downloaded successfully
4. CSV file format matches requirements

---

Last Updated: 2026-02-06
Version: 2.0.0 (Modular Architecture)