Skip to content

README.md

Latest commit

 

History

History
233 lines (161 loc) · 4.94 KB

README.md

File metadata and controls

233 lines (161 loc) · 4.94 KB

LiaPlus AI – Conversational Sentiment Analysis Chatbot

A production-ready chatbot built for the LiaPlus AI internship assignment. This project provides:

  • Tier 1 (mandatory): Conversation-level sentiment analysis.
  • Tier 2 (bonus): Statement-level sentiment analysis.

This updated version uses VADER Sentiment Analyzer only, ensuring lightweight, fast, and dependency‑safe execution.

📌 Features

✅ Tier 1 — Overall Conversation Sentiment (Mandatory)

At the end of the interaction, the system generates:

  • Overall conversation sentiment (positive / neutral / negative)
  • Confidence score
  • Optional mood trend (improving / worsening / stable)

✅ Tier 2 — Per‑Message Sentiment (Bonus)

For every user message:

  • Sentiment is analyzed using VADER
  • Confidence is computed from compound score
  • Chatbot chooses a tone‑appropriate reply

Example:

User: "Your service disappoints me"
→ Sentiment: negative (confidence: 0.82)
Bot: "I’m sorry you’re facing trouble. Let me help fix this."

📌 Rule‑Based NLU (Context Awareness)

A small NLU classifier detects intent using keyword patterns.

Intent Example Keywords
greeting hi, hello, hey
farewell bye, thanks
refund refund, money back
delivery_issue late, package, not delivered
technical_issue error, crash, not working
billing_issue bill, charge, invoice
account_issue login, password
general fallback

This allows context‑specific replies:

User: "my package is late"
Bot: "I’m sorry your package is delayed. Could you share your order ID?"

📂 Project Structure

src/
├── chatbot/
│   ├── chatbot.py
│   ├── conversation_manager.py
│   ├── response_generator.py
│   └── utils.py
├── components/
│   ├── sentiment_component.py
│   ├── text_cleaner.py
│   └── intent_classifier.py
├── services/
│   ├── sentiment_service.py
│   └── conversation_service.py
├── repository/
│   └── conversation_repository.py
├── analytics/
│   └── mood_shift_detector.py
├── utils/
│   ├── logger.py
│   └── formatters.py
main.py

🚀 How to Run the Project

1️⃣ Create virtual environment

python -m venv venv

Activate it:

# Windows
venv\Scripts\activate

# Linux / Mac
source venv/bin/activate

2️⃣ Install dependencies

pip install -r requirements.txt

3️⃣ Run the chatbot (CLI)

python main.py

4️⃣ End the conversation with:

quit
exit
bye

🧠 Sentiment Logic Explained

✔ Tier 2 (Statement‑Level)

The system uses VADER only:

  • compound score determines sentiment label
  • Confidence = absolute value of compound
Compound Score Range Meaning
≥ 0.05 Positive
≤ -0.05 Negative
Between Neutral

This ensures predictable, consistent behavior.

✔ Tier 1 (Conversation‑Level)

At the end:

  • All message compound scores are averaged
  • Higher confidence weights influence final sentiment

Also computes:

  • Trend: improving / worsening / stable

🟦 Technologies Used

Core NLP

  • VADER (NLTK)
  • Rule-Based NLU
  • Text preprocessing utilities

Architecture

  • Modular components
  • Service layer
  • Repository layer for saving conversations
  • Logging utilities

Testing

  • pytest for unit tests on:

    • Text cleaner
    • Sentiment component
    • Conversation manager

🏆 Status of Tier 2 Implementation

Feature Status
Per-message sentiment ✅ Done
Confidence scoring ✅ Done
Sentiment-aware tone ✅ Done
Mood trend analysis ⭐ Bonus Done
Transformers model ❌ Removed (now uses VADER only)

💬 Example Chat Output

Bot: Hello! I'm Leoplus Assistant. How can I help?

You: My package is not delivered yet.
→ Sentiment: negative (0.81)
Bot: I’m sorry your package is delayed. Could you share your order ID?

You: Also the billing was wrong.
→ Sentiment: negative (0.73)
Bot: I apologize for the billing trouble. What seems incorrect?

quit

=== Conversation Summary ===
Overall Sentiment: negative (0.77)
Trend: worsening

🎯 Project characteristics

  • Clean, modular, production-style architecture
  • Tier 1 + Tier 2 fully satisfied
  • Rule-Based NLU improves realism
  • Stable sentiment system using VADER
  • Clear documentation
  • Lightweight (no heavy ML models needed)