Dataset Card for aslan-ng/amazon_products_2023

Dataset Summary

This dataset is a filtered and cleaned version of the original
milistu/AMAZON-Products-2023 dataset.

It contains product titles, descriptions, main categories, average ratings, and rating counts for a broad selection of Amazon listings from 2023, excluding media and digital product categories such as books, music, and software.

The goal of this dataset is to provide a compact, cleaned corpus for text-based regression and classification tasks, such as predicting product categories or modeling rating patterns from product descriptions.

Supported Tasks

• Product Category Classification — predict main_category from text.
• Rating Regression / Quality Estimation — predict average_rating or rating_number from title and description.
• Representation Learning — use product metadata for contrastive or embedding-based models.

Languages

• English (en)

Dataset Structure

Data Instances

{
  "title": "Wireless Bluetooth Earbuds, Noise Cancelling, Long Battery Life",
  "description": "True wireless earbuds with active noise cancellation, ergonomic design, and 30 hours of playtime.",
  "main_category": "meta_Electronics",
  "average_rating": 4.5,
  "rating_number": 18421,
  "product_quality_score": 4.6
}

Calculation of product_quality_score

product_quality_score is computed using a Bayesian average formula to reduce bias toward products with few ratings.
The score combines each product’s observed average (R) and the global mean (C) weighted by the number of votes (v) and a prior constant (m):

score = (v / (v + m)) × R + (m / (v + m)) × C

where:

• ( R ) = product’s average rating
• ( v ) = number of ratings
• ( C ) = mean rating across all products
• ( m ) = minimum votes threshold (empirically chosen, e.g. 100)

The resulting score is then normalized to [0, 1] for consistency across tasks.

This approach favors well-rated products that also have sufficient review volume, mitigating small-sample noise.

Data Splits

Split	Examples
train	depends on your final filtered count (insert number if known)

Curation Rationale

The source dataset (milistu/AMAZON-Products-2023) covers a very wide range of categories, many of which (like books, music, software, and games) are textually biased or less suitable for product feature extraction tasks.

Therefore, this dataset intentionally removes those categories to focus on tangible consumer goods — such as electronics, home, beauty, and tools — which have more descriptive and visually grounded features.

Data Processing

The preprocessing steps were:

1. Start from: milistu/AMAZON-Products-2023
2. Filtered out the following categories:

   main_categories_to_remove = [
       "meta_Books", "meta_CDs_and_Vinyl", "meta_Digital_Music",
       "meta_Gift_Cards", "meta_Grocery_and_Gourmet_Food",
       "meta_Magazine_Subscriptions", "meta_Software", "meta_Video_Games"
   ]
   

3. Kept only the following columns:

   cols_to_keep = ["title", "description", "main_category", "average_rating", "rating_number"]
   

4. Removed rows with missing values in any of the selected columns.
5. Saved the result as the new aslan-ng/amazon_products_2023 dataset.

Source Data

• Original dataset: milistu/AMAZON-Products-2023
• Source type: Publicly available product metadata and textual data.
• Collection year: 2023

All product texts (titles and descriptions) are originally authored by marketplace sellers or product manufacturers.

Annotations

• Annotation process: None (directly filtered and cleaned).
• Annotators: Dataset creator (Aslan Noorghasemi).
• Label reliability: Inherits from original dataset taxonomy.

Legal and Ethical Considerations

• Data origin: Commercially public product listings; no private data included.
• License: MIT (inherits from the original dataset license).
• Responsible use: Product texts may reflect marketing bias or linguistic exaggeration; do not use for factual analysis or high-stakes decisions.

Limitations

• Some categories may still have subtle overlaps (e.g., meta_Office_Products vs meta_Home_and_Kitchen).
• The dataset does not include price, brand, or image data.
• Ratings are static and not timestamped.
• Small category imbalance may persist after filtering.

Intended Uses

• Educational demonstrations of text classification or regression.
• Lightweight benchmarking for product categorization or rating prediction models.
• Feature-extraction or embedding fine-tuning tasks.

License

• Dataset license: MIT
• Source: milistu/AMAZON-Products-2023 (please verify the source dataset’s redistribution terms before rehosting).

Maintainers

• Maintainer: aslan-ng
• Contributions: Pull requests welcome for category balancing, multilingual extensions, or integration with structured product data.