Ontology Editor

A collection of tools for processing O*Net task data using LLM queries and integrating tasks into hierarchical ontology structures.

Overview

This repository contains template code for analyzing and processing ontology task data through:

• LLM-based row processing: Apply custom AI queries to each row of CSV data
• Hierarchy integration: Map classified tasks into JSON ontology structures

Files

Data Files

• 0926_onet_classifications.csv - O*Net task data with classifications (38,223 rows)
  • Contains: Task ID, Task description, Verb, Object, Normalized form, Verb Path, Synset, Object_Singularized
• 0926_hierarchy.json - Hierarchical ontology structure for task integration

Processing Scripts

• process_rows.py - Template for LLM-based CSV row processing
• map_to_json.py - Integrates classified tasks into hierarchy structure

Quick Start

Prerequisites

pip install pandas tqdm python-dotenv openai

Set up your OpenAI API key:

echo "OPENAI_API_KEY=your_key_here" > .env

Using process_rows.py

This script processes each row of your CSV with customizable LLM queries.

1. Basic Configuration

Edit the main() function to set your parameters:

config = {
    'input_file': '0926_onet_classifications.csv',  # Your CSV file
    'output_column_name': 'LLM_Analysis',           # Name for new column
    'llm_model': 'gpt-4o',                          # OpenAI model
    'batch_size': 1,                                # Save frequency
    'delay_between_calls': 1.0,                     # API rate limiting
    'max_retries': 3                                # Error handling
}

2. Customize the LLM Prompt

This is the key step! Edit the create_llm_prompt() method for your specific use case.

The method receives a pandas row and should return a dictionary with 'system' and 'user' prompts:

def create_llm_prompt(self, row: pd.Series, row_index: int) -> Dict[str, str]:
    # Access any column from your CSV
    task = row.get('Task', '')
    verb = row.get('Verb', '')
    obj = row.get('Object', '')
    
    # Create your custom prompts
    system_prompt = '''Your system instruction here...'''
    user_prompt = f"Your user prompt using {task}, {verb}, {obj}..."
    
    return {'system': system_prompt, 'user': user_prompt}

3. Example Use Cases

Task Classification:

system_prompt = '''Classify tasks into: Physical, Information, Communication, Creative'''
user_prompt = f"Classify this task: {task}"

Quality Assessment:

system_prompt = '''Rate verb-object alignment on 1-10 scale'''
user_prompt = f'Task: {task}\nVerb: {verb}\nObject: {obj}\nHow well does "{verb} {obj}" represent this task?'

Information Extraction:

system_prompt = '''Extract the primary tool or equipment mentioned in the task'''
user_prompt = f"What tool/equipment is used in: {task}"

4. Run the Script

python process_rows.py

The script will:

• ✅ Create automatic backups
• 📊 Show progress with live updates
• 💾 Save after each row (resumable)
• 🔄 Handle errors with retries
• 📈 Provide completion statistics

Using map_to_json.py

Integrates your classified task data into the ontology hierarchy.

1. Configuration

Edit the main() function:

config = {
    'input_csv_file': '0926_onet_classifications.csv',
    'hierarchy_file': '0926_hierarchy.json',
    'output_file': '0926_hierarchy_with_onet.json'
}

2. Run Integration

python map_to_json.py

This will:

• 🔄 Map tasks to hierarchy nodes based on synsets
• 📁 Create structured "(Atomic Tasks)" and "(Specializations)" sections
• 📊 Generate detailed integration report
• 💾 Output enhanced hierarchy JSON

More Features of process_rows.py

General

• Live Progress Tracking: Visual progress bar with current status
• Automatic Backups: Timestamped backups before processing
• Resume Capability: Skip already processed rows on restart
• Error Handling: Exponential backoff retry logic for API failures
• Flexible Configuration: Easy customization for different use cases
• Debug Output: Detailed logging of each processing step

Custom Column Analysis

Process specific columns by modifying the prompt creation:

# Multi-column analysis
task_id = row.get('Task ID', '')
normalized = row.get('Normalized', '')
synset = row.get('Synset', '')

user_prompt = f"""
Task ID: {task_id}
Task: {task}
Normalized: {normalized}
Synset: {synset}

Analyze the consistency between these fields...
"""

Batch Processing

For large datasets, adjust batch processing:

config = {
    'batch_size': 10,        # Save every 10 rows
    'delay_between_calls': 0.5,  # Faster processing
}

Error Recovery

If processing fails, simply restart the script - it will resume from where it left off by checking for empty cells in your output column.

Output

process_rows.py Output

• Updates your CSV with a new column containing LLM responses
• Creates timestamped backup files
• Provides processing statistics and error counts

map_to_json.py Output

• Enhanced hierarchy JSON with integrated tasks
• ONet_Integration_Report.md with detailed analysis
• Statistics on integration coverage and missing synsets

Troubleshooting

• API Errors: Check your OpenAI API key and rate limits
• File Not Found: Verify file paths in configuration
• Memory Issues: Reduce batch_size for large files
• Resume Issues: Delete problematic output column to restart fresh