feat: v2.0.0 real data - README.md

README.md

@@ -0,0 +1,236 @@
+---
+license: cc-by-nc-4.0
+language:
+  - en
+tags:
+  - ecommerce
+  - temu
+  - price-monitoring
+  - competitor-price-monitoring
+  - sku-variants
+  - product-matching
+  - dynamic-pricing
+  - web-scraping
+  - retail
+  - anti-bot
+pretty_name: Temu E-commerce Pricing & SKU Variant Dataset
+size_categories:
+  - n<1K
+task_categories:
+  - text-classification
+  - other
+task_ids:
+  - multi-label-classification
+annotations_creators:
+  - no-annotation
+multilinguality:
+  - monolingual
+source_datasets:
+  - original
+---
+
+# Temu E-commerce Pricing & SKU Variant Dataset
+
+**85 products · 382 SKUs · $2.14–$219.72 price range · avg 29.5% discount where measurable**
+
+A real, production-quality sample of Temu product listings and SKU-level pricing data, captured by **[Octoparse Managed Data Service](https://www.octoparse.com/managed-data-service)** via a managed anti-bot pipeline. Every row is real market data — no synthetic expansion, no mock prices.
+
+Built for teams working on **competitor price monitoring**, **dynamic pricing models**, **product matching**, and **e-commerce AI pipelines**.
+
+> **v2.0.0** — Complete rebuild. Previous version contained synthetic data; this release contains 100% real scraped data normalized into a clean two-table schema.
+
+---
+
+## Why This Dataset
+
+Most public e-commerce datasets are:
+- **Stale** — from Kaggle uploads years old, prices meaningless
+- **Synthetic** — generated to look like pricing data, not real market signals
+- **Flat and noisy** — video metadata and SKU variants collapsed into one bloated table
+
+This dataset is different:
+- **100% real** market data from Temu's live catalogue
+- **Normalized schema**: `products.parquet` (SPU-level) + `skus.parquet` (variant-level), joined on `spu_id`
+- **Anti-bot provenance**: captured through Akamai Bot Manager bypass and TLS fingerprint randomization — the hard infrastructure problem solved for you
+- **Pricing signal transparency**: `discount_pct` computed where Temu surfaces both prices; sparsity documented, not hidden
+
+---
+
+## Dataset Structure
+
+```
+temu-ecommerce-pricing-workflow-sample/
+├── products.parquet        ← 85 rows, 5 columns (SPU-level)
+├── skus.parquet            ← 382 rows, 14 columns (SKU-level)
+├── schema.json             ← Machine-readable field definitions
+├── data_dictionary.md      ← Business context for every field
+├── workflow_stats.json     ← Pipeline provenance and anti-bot metadata
+└── LICENSE                 ← CC BY-NC 4.0
+```
+
+### `products.parquet` — SPU Level
+
+| Field | Type | Description |
+|---|---|---|
+| `spu_id` | string | Product ID (join key) |
+| `product_url` | string | Temu product URL |
+| `is_on_sale` | boolean | Currently purchasable |
+| `not_on_sale_reason` | string | Reason if inactive (null for all active) |
+| `goods_property_summary` | string | Semicolon-delimited property bag |
+
+### `skus.parquet` — SKU Level
+
+| Field | Type | Description |
+|---|---|---|
+| `sku_id` | string | Variant ID |
+| `spu_id` | string | FK → `products.spu_id` |
+| `sku_main_pic` | string | CDN image URL |
+| `sku_original_price` | float64 | List price USD (null if Temu not showing strikethrough) |
+| `sku_discounted_price` | float64 | Current selling price USD |
+| `discount_pct` | float64 | `(original - discounted) / original` |
+| `sale_attr_color` | string | Color variant |
+| `sale_attr_size` | string | Size variant |
+| `sale_attr_quantity` | string | Bundle/quantity variant |
+| `sale_property_summary` | string | Pipe-delimited summary of all variant attrs |
+
+---
+
+## Quick Start
+
+```python
+import pandas as pd
+
+products = pd.read_parquet("hf://datasets/Octoparse/temu-ecommerce-pricing-workflow-sample/products.parquet")
+skus     = pd.read_parquet("hf://datasets/Octoparse/temu-ecommerce-pricing-workflow-sample/skus.parquet")
+
+# Join: full product + SKU view
+df = skus.merge(products[["spu_id", "product_url", "goods_property_summary"]], on="spu_id")
+
+# Products with highest discount (where data available)
+top_discounts = (
+    skus[skus["discount_pct"].notna()]
+    .groupby("spu_id")["discount_pct"]
+    .max()
+    .sort_values(ascending=False)
+    .head(10)
+)
+
+# Price spread per product (variant pricing range)
+price_spread = (
+    skus.groupby("spu_id")["sku_discounted_price"]
+    .agg(["min", "max"])
+    .assign(spread=lambda x: x["max"] - x["min"])
+    .sort_values("spread", ascending=False)
+)
+
+# SKUs with strikethrough pricing (visible discount signal)
+has_strikethrough = skus[skus["sku_original_price"].notna()].copy()
+print(f"{len(has_strikethrough)} SKUs with measurable discount — avg {has_strikethrough['discount_pct'].mean():.1%}")
+```
+
+---
+
+## Dataset Stats
+
+| Metric | Value |
+|---|---|
+| Products (SPUs) | 85 |
+| SKUs (variants) | 382 |
+| Avg SKUs per product | 4.5 |
+| SKUs with strikethrough price | 120 (31.4%) |
+| Avg discount (where measurable) | 29.5% |
+| Max discount | 81.9% |
+| Price range | $2.14 – $219.72 |
+| All products on sale | Yes (100%) |
+
+---
+
+## The Hard Problem: Why Temu Data is Difficult to Scrape
+
+Temu runs one of the most aggressive anti-bot stacks in consumer e-commerce:
+
+- **Akamai Bot Manager** with TLS/JA3 fingerprinting — blocks standard HTTP clients instantly
+- **Device fingerprinting** — canvas, WebGL, and font enumeration checks
+- **Dynamic price rendering** — prices are injected via JavaScript after page load; HTML-only scrapers return empty price fields
+- **Signed CDN URLs** — image URLs rotate signatures; naive URL caching breaks within hours
+- **Algorithmic discount visibility** — Temu controls when strikethrough prices appear; the same SKU may show or hide the original price depending on session context, geo, and time
+
+This dataset was captured through Octoparse's managed pipeline that handles all of the above. The `sku_original_price` sparsity you see (69% null) is a **real market signal**, not a data quality failure — it accurately reflects which SKUs Temu was displaying comparison pricing for at scrape time.
+
+---
+
+## Pricing Signal Notes
+
+**Why is `sku_original_price` 69% null?**
+
+Temu uses algorithmic price display — the strikethrough "was" price is only rendered when Temu's system determines it increases conversion for that session. This means:
+- Null `sku_original_price` ≠ no discount
+- Null `sku_original_price` = Temu chose not to show the comparison at this moment
+- For true discount rate measurement, you need **repeated captures** at different times and sessions
+
+For production-cadence pricing (hourly snapshots, session rotation), see our managed pipeline service.
+
+---
+
+## Use Cases
+
+### 1. Competitor Price Monitoring Pipeline Design
+Use `sku_discounted_price` + `discount_pct` as features in a price alert system. The `sale_attr_*` columns let you track price moves at variant granularity (not just product level).
+
+### 2. Product Matching & Deduplication
+`spu_id` + `sku_id` + `sku_main_pic` give you the triplet needed for visual and text-based product matching across platforms.
+
+### 3. Dynamic Pricing Model Training
+The `discount_pct` distribution (mean 29.5%, max 81.9%) and the presence/absence of `sku_original_price` are themselves signals for modeling Temu's promotional logic.
+
+### 4. E-commerce Schema Design
+Use `goods_property_summary` to build a product taxonomy. The ~178 unique property keys across 85 products show the schema complexity of a live multi-category catalogue.
+
+### 5. Anti-Bot Research
+`workflow_stats.json` documents the specific bot-management systems encountered during collection, useful for anti-scraping research and pipeline architecture planning.
+
+---
+
+## Data Limitations
+
+| Limitation | Detail |
+|---|---|
+| Price is a snapshot | No time-series in this sample. Temu prices change multiple times daily. |
+| `sku_original_price` sparsity | 69% null — intentional (see Pricing Signal Notes above) |
+| `sale_attr_*` sparsity | Each SKU uses at most 1–2 of 7 attribute columns |
+| Property key inconsistency | `goods_property_summary` has ~178 keys with typos/case variations |
+| Sample size | 85 products — sufficient for schema/pipeline design, not for statistical price benchmarking |
+
+---
+
+## Want Production-Scale Temu Data?
+
+This is a **sample dataset**. If you need:
+- **Hourly price capture** across thousands of SKUs
+- **Multi-category coverage** (apparel, electronics, home, beauty)
+- **Session-rotated scraped** to expose hidden discount pricing
+- **Delivery to Snowflake, BigQuery, or S3** in Parquet or JSONL
+
+→ **[Talk to Octoparse Managed Data Service](https://www.octoparse.com/managed-data-service)**
+
+We operate the anti-bot infrastructure so your team doesn't have to.
+
+---
+
+## Citation
+
+```bibtex
+@dataset{octoparse_temu_pricing_2025,
+  title        = {Temu E-commerce Pricing and SKU Variant Dataset},
+  author       = {{Octoparse Managed Data Service}},
+  year         = {2025},
+  publisher    = {Hugging Face},
+  url          = {https://huggingface.co/datasets/Octoparse/temu-ecommerce-pricing-workflow-sample},
+  license      = {CC BY-NC 4.0},
+  note         = {Real scraped Temu product and SKU pricing data, anti-bot pipeline provenance documented}
+}
+```
+
+---
+
+*Built by [Octoparse Managed Data Service](https://www.octoparse.com/managed-data-service) · [LinkedIn](https://www.linkedin.com/company/octoparse) · [GitHub](https://github.com/octoparse)*