product-title-optimizer

Product Title Optimizer

LLM-powered title restructuring with data integrity checks.

What It Does

• Restructures messy product titles to optimal word order
• Groups by category for consistent formatting
• Creates category-specific templates automatically
• Validates no data is lost (numbers, model codes, etc.)
• Works with OpenAI API or local LLMs (LM Studio, Ollama, etc.)

Use Cases

• Standardize inconsistent product naming conventions
• Improve title readability for better SEO and UX
• Apply consistent brand positioning (Brand first vs Product first)
• Batch process thousands of titles efficiently

Installation

pip install -r requirements.txt

Usage

With OpenAI API

# Using environment variable
export OPENAI_API_KEY=your-api-key
python product_title_optimizer.py --input products.csv --output optimized.csv

# Or pass key directly
python product_title_optimizer.py --input products.csv --api-key sk-xxx --output optimized.csv

With Local LLM (LM Studio)

python product_title_optimizer.py \
    --input products.csv \
    --base-url http://localhost:1234/v1 \
    --model local-model \
    --output optimized.csv

Custom Column Names

python product_title_optimizer.py \
    --input products.csv \
    --title-col "Product Name" \
    --category-col "Category" \
    --output optimized.csv

Arguments

Argument	Description	Default
--input, -i	Input CSV file	Required
--output, -o	Output CSV file	optimized_titles.csv
--title-col	Column for product titles	Name
--category-col	Column for categories	Categories
--api-key	OpenAI API key	OPENAI_API_KEY env var
--base-url	Custom API URL (for local LLMs)	None
--model	Model to use	gpt-4o
--batch-size	Titles per API call	20
--limit	Limit products (0 = no limit)	0

Input Format

Name,Categories
3M Speedglas 100 Welding Helmet Xterminator Design,Safety Equipment
DeWalt DCD796N 18V XR Brushless Compact Combi Drill Body Only,Power Tools

Output

Adds columns to your CSV:

• Optimized Title: The restructured title
• Is Same: Whether the title was changed
• Missing Words: Any words from original not in optimized (for QA)

How It Works

1. Groups products by category
2. Creates a category-specific template using LLM analysis
3. Optimizes each title using the template
4. Validates data integrity:
   • All numbers preserved
   • Word overlap check (80%+ required)
   • Logs any potential issues

Title Structure

Follows adapted Basic English word order:

Brand + Product Type + Model/Series + Key Features + Specifications + Additional Info

Example:

• Before: 3M Speedglas 100 Welding Helmet Xterminator Design with Welding Filter
• After: 3M Speedglas Welding Helmet - 100 Series - Xterminator Design - with Welding Filter

Data Integrity

The tool ensures:

• All numbers and measurements are preserved
• All model codes are kept
• No invented information is added
• Titles that fail validation keep their original

Using Local LLMs

Works with any OpenAI-compatible API:

• LM Studio: --base-url http://localhost:1234/v1
• Ollama: --base-url http://localhost:11434/v1
• Text Generation WebUI: --base-url http://localhost:5000/v1

Author

Lee Foot - leefoot.com