README.md

---
license: gpl-3.0
language:
- en
tags:
- recommendation
- reviews
- ecommerce
- ratings
- user-behavior
pretty_name: Amazon Reviews 2023 All-Category k-Core
size_categories:
- 100M<n<1B
---
# Dataset Card for `Amazon Reviews 2023 All-Category k-Core`

* These datasets are subsets of [Amazon reviews dataset](https://amazon-reviews-2023.github.io/), collected in 2023 by [McAuley Lab](https://cseweb.ucsd.edu/~jmcauley/). 

* It contains **all categories** of the reviews from the original dataset that have more than $k \in [5, 20]$ interactions. 

* The original dataset contains reviews in the period of May. 1996 to Sep. 2023. 

* The reviews are grouped into [25 categories](https://amazon-reviews-2023.github.io/#grouped-by-category). 

* The dataset is in `.parquet` format.

> [!NOTE]
> 
> **k-core** means that every user and every item has at least k interactions across **ALL categories combined**. 
>
> This condition may not hold within a single category.
>


## Dataset Details

### Dataset Description

The dataset contains reviews from Amazon, and it is a subset of the original dataset. The dataset is in `.parquet` format.

Please refer to the [Dataset Creation and Processing](#dataset-creation-and-processing) section for more details about the dataset.

### Dataset Structure

The repository is structured as follows:

```
amazon-2023-all-category-k-core/
  |- 5-core/
     |- 5-core.parquet                 # 5-core ratings of all categories, 3.16GB
  |- 20-core/
     |- category/
        |- Arts_Crafts_and_Sewing/
           |- ratings.parquet          # ratings of Arts, Crafts & Sewing
           |- meta.parquet             # meta data of items in Arts, Crafts & Sewing
           |- reviews.parquet          # reviews of items in Arts, Crafts & Sewing
        |- ...                         # other categories
     |- 20-core.parquet                # 20-core ratings of all categories, 1.1GB
     |- item_map.jsonl.gz              # item map, format: [{item_index:int, parent_asin:str}], 7.97MB
     |- user_map.jsonl.gz              # user map, format: [{user_index:int, user_id:str}],     29.4MB
```

## Dataset Creation and Processing

1. Merge the `ratings` from all categories of [Amazon reviews 2023 dataset](https://amazon-reviews-2023.github.io/)
2. Filter out the `ratings` that have less than $k$ interactions, where $k \in [5, 20]$.
3. Filter out the `meta` data and `reviews` of items that are not in the filtered `ratings`.
4. Save the datasets in `.parquet` format.

### Core Code Snippets

```python
# Iteratively remove all users and items with fewer than k ratings

k = 20

while True:
    user_counts = df['user_id'].value_counts()
    item_counts = df['parent_asin'].value_counts()
    
    filtered_df = df[
        df['user_id'].isin(user_counts[user_counts >= k].index) &
        df['parent_asin'].isin(item_counts[item_counts >= k].index)
    ]
    
    if len(filtered_df) == len(df):
        break
    
    df = filtered_df

# `df` or `filtered_df` would be the resulted data.
```

## Dataset Sources

The original dataset is available at [Amazon reviews dataset](https://amazon-reviews-2023.github.io/).

<!-- - **Repository:** [More Information Needed] -->
<!-- - **Paper [optional]:** [More Information Needed] -->
<!-- - **Demo [optional]:** [More Information Needed] -->

## Uses

This dataset can be used for recommendation systems, sentiment analysis, and other NLP tasks.

<!-- ## Citation [optional] -->

<!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->

<!-- **BibTeX:** -->

<!-- [More Information Needed] -->

<!-- **APA:** -->

<!-- [More Information Needed] -->

## Glossary

The glossary of the dataset is available at [Amazon Reviews#Data Fields](https://amazon-reviews-2023.github.io/#data-fields).

## Dataset Card Authors

Chenglong Ma

## Dataset Card Contact

https://huggingface.co/ChenglongMa