🤖 Review Classification Engine

Batch-optimized LLM-based automated customer review classification and sentiment detection engine using Ollama + Llama3.

---

📌 Problem

Ecommerce platforms receive thousands of customer reviews daily.

Manually reading and categorizing them into:

• Delivery Issues
• Payment Issues
• Product Quality
• Refund Problems

is inefficient and time-consuming.

---

🚀 Solution

This project implements a batch-optimized LLM pipeline that:

• Loads customer reviews from CSV
• Processes reviews in batches
• Classifies review category
• Detects sentiment (Positive / Neutral / Negative)
• Returns structured JSON output
• Handles parsing safely
• Includes error handling

---

🛠 Tech Stack

• Python
• Pandas
• Ollama
• Llama3
• Batch Prompt Engineering
• JSON Parsing

---

⚙️ How It Works

1. Reviews are loaded from CSV
2. Reviews are grouped into batches (default: 10 per call)
3. Llama3 processes multiple reviews in a single API call
4. Output is strictly enforced in JSON format
5. Safe JSON extraction prevents crashes
6. Errors are handled gracefully

---

📦 Installation

Install dependencies:

pip install -r requirements.txt

Make sure Ollama is installed and running:

ollama run llama3

---

▶️ Run the Script

python llm_review_batch_analyzer.py

---

📊 Example Output

[
  {"review_number": 1, "category": "Delivery Issue", "sentiment": "Negative"},
  {"review_number": 2, "category": "Payment Issue", "sentiment": "Negative"}
]

---

🔥 Performance Optimization

Instead of making 1 API call per review:

Old Approach: 30 reviews = 30 API calls ❌

New Batch Approach: 30 reviews = 3 API calls ✅

• Reduced latency
• Reduced token usage
• Scalable architecture

---

📌 Future Enhancements

• Negative sentiment alert threshold
• Save output to CSV
• Real-time dashboard integration
• Deploy as REST API