Fashion Product Data Collection — ZOZOTOWN (Japan)

Client Overview

The client required a complete, analysis-ready fashion catalog dataset from ZOZOTOWN — one of Japan's largest online fashion retail platforms — to support catalog analysis, assortment evaluation, and competitive benchmarking in the Japanese fashion retail sector.

Actowiz Solutions executed a once-off, full-catalog extraction across all available categories, capturing not just core product details but every available color and size variation for each item. Two structural requirements were central: complete variant-level coverage — where every color/size combination is captured as its own structured record — and a unique hash-based ASIN generated from Goods ID, color, and size to reliably identify each variant downstream.

Data was captured in both Japanese and English exactly as displayed on the platform, with no translation applied, preserving source-language fidelity. The final dataset was delivered in Excel/CSV — clean, deduplicated, and structured for immediate analytical use.

The Challenge

• Variant-level complexity. Fashion products commonly carry multiple color and size combinations per listing. Capturing every combination as a distinct, correctly mapped record — rather than collapsing variants into a single row — significantly increased extraction complexity.
• Reliable variant identification. The platform's native Goods ID (used as PID) identifies a product but not a specific variant. A consistent method was needed to uniquely identify each color/size combination without relying on platform-native variant IDs alone.
• Bilingual content without translation. Product names and descriptions had to be captured in whichever language the platform displayed (Japanese or English), with strict instructions not to apply any translation layer that could alter source meaning.
• Pricing logic consistency. MRP, sale price, final price, and discount needed strict logical consistency (e.g., sale price must not exceed MRP), requiring careful handling of the platform's pricing display across regular and discounted listings.
• Text quality. Descriptions and names had to be free of HTML tags and junk characters, requiring a dedicated text-cleaning layer applied consistently across all bilingual content.

The Solution by Actowiz Solutions

Actowiz designed a full-catalog extraction pipeline built around three core capabilities — complete category traversal, row-wise variant expansion, and hash-based variant identification — delivered as a clean, structured Excel/CSV dataset.

Variant-Level Extraction

For every product page, the pipeline captured all available color and size combinations and represented each as a separate, fully-attributed row. This row-wise expansion enables analysis at the true SKU/variant level rather than an aggregated product level — essential for accurate assortment and pricing analysis in fashion. The Goods ID (used as PID) stays consistent across all variant rows of the same parent product, so variants can be grouped back to their parent whenever needed.

Hash-Based ASIN Generation

To uniquely identify each variant, a hash-based ASIN is generated by combining the product's Goods ID with its specific Color and Size values. This produces a deterministic, unique identifier for every color/size combination — the same variant always resolves to the same ASIN — supporting consistent variant tracking without depending on any platform-native variant ID.

Bilingual Content Capture

Names, descriptions, and other text were extracted exactly as displayed — Japanese, English, or a mix — depending on what the platform presented for each product. No translation layer was applied at any stage, preserving full fidelity while still passing through the text-cleaning layer that removes HTML tags and junk characters.

Pricing Logic & Standardization

MRP, sale price, and final price were cross-checked for logical consistency — sale price validated as ≤ MRP, with discount calculated as the numeric difference between MRP and final price. All prices were standardized to Japanese Yen (JPY), and boolean fields such as IsOnSale and IsInStock normalized to consistent TRUE/FALSE values.

Technical Approach

• Category traversal engine. Systematic navigation across all accessible ZOZOTOWN categories, supplemented by manual coverage verification due to the absence of a reliable sitemap.
• PDP parsing. Structured extraction of all core attributes from each product detail page — description, pricing, brand, and image data.
• Variant expansion logic. Automated detection and row-wise expansion of all color/size combinations, so every variant is its own complete record.
• Hash-based ASIN generation. Deterministic hashing of Goods ID + Color + Size to produce a stable, unique variant identifier.
• Bilingual text capture. Language-agnostic extraction capturing whatever language ZOZOTOWN displays per field, with no translation logic applied.
• Text-cleaning pipeline. HTML tag stripping and junk-character removal applied uniformly across all descriptive fields, regardless of source language.
• Pricing validation logic. Automated cross-checks ensuring Sale Price ≤ MRP and Rating ≤ 5, with calculated Discount and standardized JPY tagging.
• Image URL handling. Main and alternate images extracted separately; alternates concatenated with a pipe delimiter.
• Deduplication & schema enforcement. Final dataset deduplicated and validated against the fixed 28-column schema before delivery.

Results & Business Impact

• Complete catalog visibility. Full category coverage gave clear visibility into the breadth of ZOZOTOWN's assortment, supporting catalog benchmarking; breadcrumb paths enabled category-level rollups at any hierarchy level.
• True variant-level analysis. Row-wise color/size expansion enabled accurate variant-level assortment analysis — size-curve and color-mix analysis across the catalog — with hash-based ASINs providing a stable, reusable variant identifier.
• Pricing & promotion intelligence. Consistent MRP, sale, final, and discount fields gave clear visibility into promotional activity and price positioning, standardized to JPY.
• Multilingual data fidelity. Capturing data in original Japanese or English without translation preserved source fidelity for any downstream language-specific processing.
• Analysis-ready delivery. Clean, deduplicated, schema-consistent Excel/CSV required no further transformation, loading directly into analytical tools within the 3–4 working-day window.

Client Feedback

"Every color and size as its own row with a stable ID meant real variant-level analysis at last. Bilingual content preserved, zero cleanup — it loaded straight into our tools."

— Lead Catalog Analyst

Need a custom data pipeline for your platform?

Actowiz Solutions designs custom, large-scale scraping, extraction, and API-delivery pipelines with rigorous QA. Visit actowizsolutions.com to discuss your data requirement.