gilberty005/personas-product-database

Personas Product Database

Product database for running recommendation experiments with the Personas framework.

Dataset Summary

This dataset contains product information and categorization data used for conversational recommendation experiments.

Statistics:

• Products: 1,279,364
• Categories: 7,397
• Product-Category Links: 2,636,558
• Cached Persona Scores: 0

Files

• products.parquet - Product information (id, title, price, category, raw metadata)
• categories.parquet - Category names and IDs
• product_category.parquet - Many-to-many links between products and categories
• persona_scores.parquet - Cached LLM scores for persona-product pairs
• metadata.json - Export metadata and statistics

Schema

products

Column	Type	Description
id	int	Internal product ID
external_id	str	Amazon ASIN
title	str	Product name
main_category	str	Amazon's main category
store	str	Store name
price	float	Price in USD
raw	str	Full product metadata (JSON string)

categories

Column	Type	Description
id	int	Category ID
name	str	Category name

product_category

Column	Type	Description
product_id	int	References products.id
category_id	int	References categories.id

persona_scores

Column	Type	Description
persona_index	int	Persona ID (from Tianyi-Lab/Personas)
category_id	int	References categories.id
product_id	int	References products.id
score	float	Persona's score for product (0-100)
reason	str	Explanation for score
model	str	LLM model used for scoring
created_at	int	Unix timestamp

Usage

Download with Python

from datasets import load_dataset

# Load as HuggingFace dataset
dataset = load_dataset("gilberty005/personas-product-database")

# Access individual tables
products = dataset['products'].to_pandas()
categories = dataset['categories'].to_pandas()

Rebuild SQLite Database

# The experiment code includes auto-download and rebuild functionality
# Just run your experiment and the database will be created automatically!

from pipeline.core.simulate_interaction import ensure_database
ensure_database()  # Downloads Parquet files and builds SQLite DB

Related Datasets

• Personas: Tianyi-Lab/Personas - 200K synthetic customer personas

Version

• Version: 1.0
• Export Date: 2025-10-30T13:53:35.983717

License

MIT License