MightyOctopus/amazon-pricer-dataset-v2-0

Amazon Pricer Dataset v2.0

Check The Data Curation Processes

1. Data Curation & Cleaning: https://github.com/mightyoctopus/fine-tuning-frontier-llm-models/blob/main/d2_data_curation_part2.ipynb
2. Feature Engineering & Performance Tests with ML Models: https://github.com/mightyoctopus/fine-tuning-frontier-llm-models/blob/main/d3_part3.ipynb
3. Fine Tuning a GPT model with the dataset: https://github.com/mightyoctopus/fine-tuning-frontier-llm-models/blob/main/d5_part5.ipynb

Dataset Description

This dataset contains structured natural language prompts paired with ground-truth product prices derived from the McAuley-Lab/Amazon-Reviews-2023 dataset.

Unlike raw product listings, the prompt text in each sample is programmatically generated from engineered product features. Product attributes are first extracted and normalized through feature engineering, and then an LLM is used to synthesize these attributes into a consistent natural language description.

Version v2.0 extends v1 by introducing explicit Main_Category information and a more structured prompt template, improving input consistency for language model training.

The dataset is designed for text-to-number regression tasks, such as predicting product prices from product descriptions.

Dataset Structure

Each training sample follows a structured prompt format designed for price regression.

Prompt Template

• How much does this cost to the nearest dollar?

• Main_Category: {product_category}

• {LLM-generated product description} (Based on the product features extracted)

The product description is generated from structured attributes extracted from the raw dataset, rather than directly copied from the original listing text.

Feature Engineering Pipeline

Each sample is created through the following steps:

1. Extract structured product attributes from the source dataset

2. Normalize and clean feature fields (category, product attributes, metadata)

3. Construct a feature representation of the product

4. Generate a natural-language description from these features using an LLM

5. Combine the generated description with a standardized pricing prompt

Data Fields

• text: Structured natural-language prompt containing product description and category
• price: Ground-truth product price (USD)

Source

McAuley-Lab/Amazon-Reviews-2023

Intended Use

• LLM-based price prediction
• Benchmarking against classical ML models
• Research and educational purposes

Out-of-Scope Use

• Real-time pricing systems
• Financial or commercial decision-making

License

MIT