What are the best product datasets for ecommerce analysis?

The best dataset for e-commerce analysis depends on your specific goal—whether you are working on demand forecasting, recommendation engines, customer segmentation, search relevance, or sentiment analysis.

---

1. Sales, Transactions & Customer Segmentation

Best for: Customer Lifetime Value (CLV), RFM (Recency, Frequency, Monetary) analysis, churn modeling, and market basket analysis.

• Brazilian E-Commerce Public Dataset by Olist (Kaggle)
  • What it contains: ~100,000 anonymized real orders placed on Olist (a Brazilian marketplace) between 2016 and 2018. It includes 9 relational tables: order status, price, payment details, freight performance, customer location, product attributes, and customer review scores.
  • Best for: Complex relational SQL queries, order logistics/delivery performance, customer churn, and cross-category sales behavior.
• UCI Online Retail Dataset
  • What it contains: ~540,000 transnational purchase records from a UK-based non-store gift retailer (2010–2011).
  • Best for: RFM analysis, cohort analysis, and market basket association rules (e.g., Apriori algorithm). (Note: Ideal for learning/practice, though too dated for real-world time-series forecasting).
• Instacart Market Basket Analysis (Kaggle)
  • What it contains: Over 3 million anonymized order records from over 200,000 Instacart users.
  • Best for: Sequence prediction (predicting what products a user will reorder in their next cart) and product-association mining.

---

2. Product Catalogs, Recommendations & Clickstreams

Best for: Collaborative filtering, content-based filtering, session tracking, and graph networks.

• Amazon Product Data (Julian McAuley / AWS / Hugging Face)
  • What it contains: Massive datasets containing product metadata (descriptions, price, brand, categories, image links), co-purchasing links ("bought together"), and ratings across millions of products. Updated iterations (e.g., Amazon Reviews 2023) feature over 500M reviews.
  • Best for: Large-scale recommendation systems, graph neural networks, and text/embedding analysis.
• OTTO – Multi-Objective Recommender Dataset (Kaggle)
  • What it contains: Real-world clickstream session data from OTTO (Germany’s largest online retailer), tracking clicks, cart additions, and order events.
  • Best for: Real-time session-based recommendations and conversion predictions.

---

3. Product Search & E-Commerce Relevance

Best for: Information Retrieval (IR), search relevance ranking, query understanding, and Semantic Search.

• Amazon ESCI (Shopping Queries Dataset)
  • What it contains: Over 1 million query-product pairs with human annotations indicating match relevance categorized as Exact, Substitute, Complement, or Irrelevant (ESCI).
  • Best for: Fine-tuning e-commerce search engines, learning-to-rank (LTR) algorithms, and vector search embeddings.

---

4. Demand Forecasting & Pricing Analysis

Best for: Time-series forecasting, dynamic pricing models, and stock optimization.

• M5 Forecasting Dataset / Walmart (Kaggle)
  • What it contains: Hierarchical daily unit sales data for 3,049 products sold across 10 Walmart stores in the US, along with promotional calendars, prices, and special event tags.
  • Best for: Advanced time-series forecasting (XGBoost, Prophet, LSTM, N-BEATS), analyzing price elasticity, and predicting stockouts.

---

5. Product Reviews & Customer Sentiment (NLP)

Best for: Aspect-based sentiment analysis, topic modeling, and natural language processing.

• Women’s E-Commerce Clothing Reviews (Kaggle)
  • What it contains: ~23,000 real customer reviews and ratings on women’s apparel, including recommendation tags, review text, customer age, and clothing category.
  • Best for: Aspect-based sentiment analysis (detecting if complaints relate to sizing, fabric quality, or shipping speed).
• Multi-Domain Sentiment Dataset
  • What it contains: Amazon product reviews across various categories (Books, Electronics, Kitchen, etc.) labeled with binary/star sentiment.
  • Best for: Training NLP sentiment classifiers across different product verticals.

---

6. Computer Vision & Visual Search

Best for: Product image classification, visual search, tag generation, and attribute extraction.

• DeepFashion & DeepFashion2
  • What it contains: Over 800,000 high-resolution fashion images annotated with detailed attributes (pattern, fabric, sleeve style), keypoints, and cross-domain product matching links.
  • Best for: "Shop the Look" visual search, automatic product attribute tagging, and image similarity retrieval.

---

7. Commercial & Live Data Providers

If you require live e-commerce data (e.g., dynamic daily pricing across competitors like Amazon, Target, Walmart, or eBay) rather than static historical training datasets, public datasets become outdated quickly. Enterprise data providers specialize in real-time, scraped product catalogs:

• Bright Data – Offers pre-collected datasets covering 260+ e-commerce marketplaces with daily price/inventory tracking.
• Oxylabs & Grepsr – Provide web-scraping APIs and custom enterprise feeds for dynamic price, stock, and product catalog monitoring.

---

Summary Checklist: Which dataset should you pick?

Objective	Recommended Dataset
Sales Analytics & SQL Portfolio	Olist Public Dataset or UCI Online Retail
Market Basket / Cross-Selling	Instacart Market Basket
Recommendation Engine	Amazon Product Data or OTTO Clickstream
Demand / Inventory Forecasting	M5 Walmart Dataset
Search Engine / Ranking AI	Amazon ESCI Dataset
Review Sentiment (NLP)	Women's Clothing Reviews or Amazon Reviews 2023
Visual Search / Image ML	DeepFashion2