# Initial Product Corpus

(Initial because somethings may change in the future)

This dataset contains a **combined and deduplicated product metadata corpus**, created by merging two sources:

- [**Amazon ESCI Dataset**](https://github.com/amazon-science/esci-data)
- [**Amazon M2 Dataset**](https://arxiv.org/pdf/2307.09688)

The merged dataset contains product records that match a corresponding entry in the [Amazon Reviews 2023](https://huggingface.co/datasets/McAuley-Lab/Amazon-Reviews-2023) dataset (based on product ID). While **review data is not included**, users can externally enrich each product using this linkage.

---

## 📦 Dataset Format

- Format: **JSON Lines (`.jsonl.gz`)**
- Each line is a single product record as a JSON object.
- Missing values are represented as `null`.

---

## 🧾 Column Descriptions

| Column                | Description                                                   |
|-----------------------|---------------------------------------------------------------|
| `id`                  | Unique product identifier (e.g., ASIN)                        |
| `locale`              | Region or language variant (e.g., `us`, `uk`)                 |
| `title`               | Product title                                                 |
| `desc`                | Full product description                                      |
| `price`               | Product price in float format (if available)                 |
| `brand`               | Brand name                                                    |
| `color`               | Color of the product                                          |
| `size`                | Product size (especially for apparel)                         |
| `model`               | Model name or number                                          |
| `material`            | Product material (e.g., cotton, metal)                        |
| `author`              | Author name (for books, music, etc.)                          |
| `product_bullet_point`| Marketing bullet points from the listing                      |

---

## 📊 Schema & Sparsity

| Column                | Non-Null Count | Non-Null % | Type     |
|-----------------------|----------------|------------|----------|
| id                    | 518,970        | 100.0%     | string   |
| locale                | 518,970        | 100.0%     | string   |
| title                 | 518,970        | 100.0%     | string   |
| desc                  | 291,278        | 56.1%      | string   |
| price                 | 47,374         | 9.1%       | float    |
| brand                 | 480,665        | 92.6%      | string   |
| color                 | 272,382        | 52.5%      | string   |
| size                  | 22,340         | 4.3%       | string   |
| model                 | 28,629         | 5.5%       | string   |
| material              | 21,845         | 4.2%       | string   |
| author                | 8,569          | 1.7%       | string   |
| product_bullet_point  | 376,224        | 72.5%      | string   |

---

## 🧪 Example Record

```json
{
  "id": "B08K2S1VJ6",
  "locale": "us",
  "title": "Wireless Bluetooth Earbuds",
  "desc": "Experience high-fidelity sound with our latest Bluetooth 5.0 earbuds...",
  "price": 29.99,
  "brand": "SoundMax",
  "color": "Black",
  "size": null,
  "model": "SM-BT2021",
  "material": "Plastic",
  "author": null,
  "product_bullet_point": "Noise cancellation; Long battery life; Compact design"
}
````

---

## 📁 File Contents

* `data.jsonl.gz`: The complete corpus (519k records)
* `README.md`: This documentation file