Amazon Reviews 2023

Please also visit amazon-reviews-2023.github.io/ for more details, loading scripts, and preprocessed benchmark files.

[April 7, 2024] We add two useful files:

1. all_categories.txt: 34 lines (33 categories + "Unknown"), each line contains a category name.
2. asin2category.json: A mapping between parent_asin (item ID) to its corresponding category name.

---

This is a large-scale Amazon Reviews dataset, collected in 2023 by McAuley Lab, and it includes rich features such as:

1. User Reviews (ratings, text, helpfulness votes, etc.);
2. Item Metadata (descriptions, price, raw image, etc.);
3. Links (user-item / bought together graphs).

What's New?

In the Amazon Reviews'23, we provide:

1. Larger Dataset: We collected 571.54M reviews, 245.2% larger than the last version;
2. Newer Interactions: Current interactions range from May. 1996 to Sep. 2023;
3. Richer Metadata: More descriptive features in item metadata;
4. Fine-grained Timestamp: Interaction timestamp at the second or finer level;
5. Cleaner Processing: Cleaner item metadata than previous versions;
6. Standard Splitting: Standard data splits to encourage RecSys benchmarking.

Basic Statistics

We define the #R_Tokens as the number of tokens in user reviews and #M_Tokens as the number of tokens if treating the dictionaries of item attributes as strings. We emphasize them as important statistics in the era of LLMs.

We count the number of items based on user reviews rather than item metadata files. Note that some items lack metadata.

Compared to Previous Versions

Grouped by Category

Check Pure ID files and corresponding data splitting strategies in Common Data Processing section.

Quick Start

Load User Reviews

from datasets import load_dataset

dataset = load_dataset("McAuley-Lab/Amazon-Reviews-2023", "raw_review_All_Beauty", trust_remote_code=True)
print(dataset["full"][0])

{'rating': 5.0,
 'title': 'Such a lovely scent but not overpowering.',
 'text': "This spray is really nice. It smells really good, goes on really fine, and does the trick. I will say it feels like you need a lot of it though to get the texture I want. I have a lot of hair, medium thickness. I am comparing to other brands with yucky chemicals so I'm gonna stick with this. Try it!",
 'images': [],
 'asin': 'B00YQ6X8EO',
 'parent_asin': 'B00YQ6X8EO',
 'user_id': 'AGKHLEW2SOWHNMFQIJGBECAF7INQ',
 'timestamp': 1588687728923,
 'helpful_vote': 0,
 'verified_purchase': True}

Load Item Metadata

dataset = load_dataset("McAuley-Lab/Amazon-Reviews-2023", "raw_meta_All_Beauty", split="full", trust_remote_code=True)
print(dataset[0])

{'main_category': 'All Beauty',
 'title': 'Howard LC0008 Leather Conditioner, 8-Ounce (4-Pack)',
 'average_rating': 4.8,
 'rating_number': 10,
 'features': [],
 'description': [],
 'price': 'None',
 'images': {'hi_res': [None,
   'https://m.media-amazon.com/images/I/71i77AuI9xL._SL1500_.jpg'],
  'large': ['https://m.media-amazon.com/images/I/41qfjSfqNyL.jpg',
   'https://m.media-amazon.com/images/I/41w2yznfuZL.jpg'],
  'thumb': ['https://m.media-amazon.com/images/I/41qfjSfqNyL._SS40_.jpg',
   'https://m.media-amazon.com/images/I/41w2yznfuZL._SS40_.jpg'],
  'variant': ['MAIN', 'PT01']},
 'videos': {'title': [], 'url': [], 'user_id': []},
 'store': 'Howard Products',
 'categories': [],
 'details': '{"Package Dimensions": "7.1 x 5.5 x 3 inches; 2.38 Pounds", "UPC": "617390882781"}',
 'parent_asin': 'B01CUPMQZE',
 'bought_together': None,
 'subtitle': None,
 'author': None}

Check data loading examples and Huggingface datasets APIs in Common Data Loading section.

Data Fields

For User Reviews

For Item Metadata

Citation

@article{hou2024bridging,
  title={Bridging Language and Items for Retrieval and Recommendation},
  author={Hou, Yupeng and Li, Jiacheng and He, Zhankui and Yan, An and Chen, Xiusi and McAuley, Julian},
  journal={arXiv preprint arXiv:2403.03952},
  year={2024}
}

Contact Us

• Report Bugs: To report bugs in the dataset, please file an issue on our GitHub.

• Others: For research collaborations or other questions, please email yphou AT ucsd.edu.