VOC Auto-Tagging System

📋 Overview

The VOC Auto-Tagging System is an intelligent tool designed to analyze large-scale user feedback. It automatically identifies key touchpoints in the user journey, issue types, and sentiment polarity, helping teams gain actionable insights efficiently.

✨ Key Features

• Smart Tagging: Automatically detects user journey stages, issue types, and sentiment.
• Batch Processing: Supports bulk data import via Excel or CSV.
• Visualization Reports: Generates visual dashboards and summary reports.
• User-Friendly Interface: No technical background required to operate.

---

🎯 Tagging Framework

User Journey Touchpoints (multi-label)

• Pre-Purchase: Brand perception, website/App experience, in-store inquiry/test drive
• Purchase Process: Ordering flow, sales service attitude, pricing transparency
• Delivery: Handover speed/process, onboarding training, store environment
• Driving Experience: Acceleration, braking, handling, comfort, noise
• Smart Features: Navigation, voice assistant, HUD/control interaction, OTA
• Charging & Energy: Home charging, public stations, range performance
• After-Sales Service: Maintenance, customer support, trade-in

Issue Types (multi-label)

• Stability | Performance | Usability | Compatibility
• Aesthetics | Interaction Logic | Safety | Service Experience | Expectation Gap

Sentiment Polarity (single-label)

• Positive | Neutral | Negative

---

🚀 Quick Start

System Requirements

• Windows 10 or later
• Recommended: 8GB+ RAM
• Stable internet connection

Launch Instructions

Method 1 (Recommended):

• Double-click voc_launcher.ps1 in the project root.

Method 2:

• Right-click voc_launcher.ps1 → Run with PowerShell.

On first run, the system will auto-install required environments.

---

💡 Usage Guide

Choose Mode on Startup

🎯 VOC Analyzer v3.0 - Main Menu

1. 🎪 Demo Mode (Quick preview)
2. 🔤 Interactive Mode (Recommended)
3. 🎓 Train New Model
4. 📊 Batch Process Files
5. 📈 Model Performance Evaluation
6. 🔄 System Initialization
7. ❌ Exit

Single Text Analysis

Go to Interactive Mode → Analyze a Single Text.

Example input:

"The voice assistant of Li Auto ONE is accurate, but the navigation sometimes takes a longer route."

Output:

• Touchpoints: Smart Navigation, Voice Assistant
• Issues: Usability
• Sentiment: Neutral
• Confidence Scores Included

---

📊 Batch File Processing

Step 1: Prepare Input File

CSV or Excel format:

text,notes  
"The voice assistant is accurate, but the navigation is off-route.", Feedback 1  
"Charging is slower than advertised.", Feedback 2  
"Sales service was great, smooth delivery process.", Feedback 3  

Step 2: Run Batch Mode

Choose Batch Process Files and follow prompts:

• File Path
• Text Column Name (default: text)
• Generate Report? (recommended: Yes)

Step 3: View Results

• Tagged CSV output
• HTML report with visual analytics

---

🎓 Model Training

To train with custom data:

Format Example:

text,touchpoints,issue_types,sentiment  
"Navigation often takes longer routes, voice not recognized","Smart Navigation,Voice Assistant","Stability,Usability",Negative  
"Delivery staff gave detailed instructions, seat massage was impressive","Delivery Training,Suspension Comfort",,Positive  

Notes:

• Use commas to separate multiple labels
• Min. 500 labeled samples recommended
• Ensure clean, accurate labeling

---

📁 File Structure

voc-analyzer/
├── voc_launcher.ps1           # Launcher script
├── README.md                  # Documentation
├── src/                       # Core source files
├── data/
│   ├── voc_sample_data.csv    # Sample data
│   ├── models/                # Trained models
│   ├── logs/                  # System logs
│   └── reports/               # Generated reports
└── config/                    # Configuration files

---

🛠️ Troubleshooting

Startup Issues

Problem: Cannot launch

• Confirm Windows 10+
• Right-click → "Run with PowerShell"

Problem: Execution policy error

• Run PowerShell as Admin: Set-ExecutionPolicy RemoteSigned -Scope CurrentUser Confirm with Y, then relaunch.

Usage Issues

File read error

• Ensure file is CSV/XLSX and not open elsewhere
• File must be UTF-8 encoded

Inaccurate tagging

• Ensure clear and complete input text
• Consider retraining the model for domain-specific language

Slow performance

• First run may download models
• For large datasets, process in batches
• Check your network connection

---

📋 Output Details

CSV Output

• Original text
• Predicted touchpoints
• Predicted issue types
• Sentiment polarity
• Confidence scores

HTML Report

• Summary stats
• Charts by category
• Correlation heatmaps
• Quality diagnostics

---

✅ Best Practices

Data Preparation

• Ensure clean, well-written text
• One feedback per line
• Remove irrelevant symbols or formatting

Batch Processing

• Recommend ≤5000 entries per batch
• Split larger files for stability
• Clear temp files regularly

Interpreting Results

• Focus on high-confidence results
• Review predictions below 0.5 manually
• Retrain periodically with new data

---

Last updated: May 26, 2025

---