amazon-reviews-2023-with-asin
收藏资源简介：
Amazon Reviews 2023 (with ASIN) 是一个为自然语言处理（NLP）任务优化的超大规模亚马逊产品评论数据集。它基于 McAuley-Lab/Amazon-Reviews-2023 原始数据集，经过精简和标准化处理，移除了与核心NLP任务关联度较低的多余字段，仅保留四个关键字段：`rating`（1.0至5.0的浮点数评分）、`title`（评论标题）、`text`（评论文本正文）以及新增的 `parent_asin`（父产品ID）。`parent_asin` 用于唯一标识和分组同一产品的不同变体（如颜色或尺寸），支持产品级分析。数据集以标准的Parquet格式存储，可通过HuggingFace `datasets` 库直接加载，无需特殊脚本。数据覆盖亚马逊34个主要产品类别，包括电子产品、家居厨房、图书、服装鞋履等，总计包含约5.71亿条英文评论。适用于多种NLP研究和应用，例如基于 `parent_asin` 和评论文本的产品分类、基于评论标题和正文的情感分析或评分预测、利用产品标识符构建推荐系统、产品评论文本摘要生成与评估，以及作为大规模英文文本语料库用于通用NLP研究。
Amazon Reviews 2023 (with ASIN) is an ultra-large-scale Amazon product review dataset optimized for natural language processing (NLP) tasks. Derived from the original McAuley-Lab/Amazon-Reviews-2023 dataset, it has undergone refinement and standardization processing: redundant fields with low relevance to core NLP tasks have been removed, and only four key fields are retained: `rating` (a floating-point rating ranging from 1.0 to 5.0), `title` (review title), `text` (main body of review text), and the newly added `parent_asin` (parent product ID). The `parent_asin` field is used to uniquely identify and group different variants of the same product (such as color or size), enabling product-level analysis. The dataset is stored in the standard Parquet format and can be directly loaded via the HuggingFace `datasets` library without requiring special scripts. It covers 34 major product categories on Amazon, including electronics, home & kitchen, books, clothing, footwear and more, containing a total of approximately 571 million English reviews. This dataset is applicable to a variety of NLP research and applications, such as product classification based on `parent_asin` and review texts, sentiment analysis or rating prediction based on review titles and main bodies, construction of recommendation systems using product identifiers, generation and evaluation of product review text summaries, and serving as a large-scale English text corpus for general NLP research.
数据集概述
Amazon Reviews 2023 (with ASIN) 是 McAuley-Lab/Amazon-Reviews-2023 的精简版本，专注于 NLP 任务，并通过 parent_asin 字段提供明确的产品标识。
与原版的区别
- 仅保留 4 个字段：
rating、title、text和parent_asin。 - 采用标准 Parquet 格式存储，无需加载脚本。
- 可直接使用
load_dataset加载。
数据结构
每条记录包含以下字段：
| 字段 | 类型 | 描述 |
|---|---|---|
rating |
float32 | 产品评分，范围 1.0 至 5.0 |
title |
string | 用户评论标题 |
text |
string | 用户评论全文 |
parent_asin |
string | 父产品 ID，用于产品级别分组与识别 |
注意：不同颜色、尺寸或款式的产品通常共享相同的
parent_asin，该字段适用于产品级分组。
数据规模
共约 5.71 亿条评论，涵盖 34 个类别，其中主要类别包括：
| 类别 | 评论数量 |
|---|---|
| All_Beauty | 701.5K |
| Amazon_Fashion | 2.5M |
| Appliances | 2.1M |
| Arts_Crafts_and_Sewing | 9.0M |
| Automotive | 20.0M |
| Baby_Products | 6.0M |
| Beauty_and_Personal_Care | 23.9M |
| Books | 29.5M |
| CDs_and_Vinyl | 4.8M |
| Cell_Phones_and_Accessories | 20.8M |
| Clothing_Shoes_and_Jewelry | 66.0M |
| Digital_Music | 130.4K |
| Electronics | 43.9M |
| Gift_Cards | 152.4K |
| Grocery_and_Gourmet_Food | 14.3M |
| Handmade_Products | 664.2K |
| Health_and_Household | 25.6M |
| Health_and_Personal_Care | 494.1K |
| Home_and_Kitchen | 67.4M |
| Industrial_and_Scientific | 5.2M |
| Kindle_Store | 25.6M |
| Magazine_Subscriptions | 71.5K |
| Movies_and_TV | 17.3M |
| Musical_Instruments | 3.0M |
| Office_Products | 12.8M |
| Patio_Lawn_and_Garden | 16.5M |
| Pet_Supplies | 16.8M |
| Software | 4.9M |
| Sports_and_Outdoors | 19.6M |
| Subscription_Boxes | 16.2K |
| Tools_and_Home_Improvement | 27.0M |
| Toys_and_Games | 16.3M |
| Video_Games | 4.6M |
| Unknown | 63.8M |
加载方式
python from datasets import load_dataset
加载特定类别
dataset = load_dataset("BarbaDLuca/amazon-reviews-2023-with-asin", data_dir="Electronics", split="train")
加载所有类别
dataset = load_dataset("BarbaDLuca/amazon-reviews-2023-with-asin", split="train")
预期用途
- 产品分类：利用
parent_asin和评论文本进行产品分组与分类。 - 情感分析：根据评论标题和文本预测评分或情感倾向。
- 推荐系统：基于产品标识构建基于物品或用户-物品的模型。
- 摘要生成：为每个产品生成或评估评论摘要。
- 通用 NLP 研究：大规模英文文本语料库，带有结构化元数据。
参考文献
该数据集源自 McAuley Lab (UCSD) 的 Amazon Reviews 2023 数据集，引用请使用：
bibtex @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} }
