General info
Users enter a free-text Product Title (e.g. “Hermès Birkin 30 Etoupe Gold”). We use rule-based logic against our Airtable lists to auto-fill Brand, Model+Size, Color, and Material fields. We prioritize exact matches – no fuzzy corrections. Any attribute not matched stays blank for manual selection; all dropdowns include an Other option.
Our Airtable source includes multilingual variants (English, German, French) from our Weglot data, so titles in any of these languages map back to a single English canonical value.
This story covers the parsing logic only; dropdown UI behavior is in a separate story.
Rules & Details
-
Exact Matching with Exploration
-
Normalize title (lowercase, strip punctuation, normalize accents).
-
By default, match only full tokens or exact phrases from Airtable.
-
Explore optional high-confidence fuzzy or proximity matching (e.g., Levenshtein distance) for edge cases – but avoid guessing at the cost of accuracy.
-
-
Attribute Dependencies
-
Model depends on detected Brand.
-
Size depends on detected Model (e.g., Birkin→sizes 25,30,35).
-
Color scoped to Brand (brand‑specific names first, then generic).
-
Material scoped to Brand (Hermès leathers vs. generic types).
-
Hardware parsing: detect terms like “Palladium”, “Gold‑plated”, “Gold Tone” – but be cautious since “Gold” may refer to either color or metal finish.
-
-
Translations & Canonical Values
-
Airtable stores EN/DE/FR variants from Weglot; match any variant and save as the English canonical term.
-
-
Other Fallback
-
If an attribute isn’t matched (or match confidence is low), leave it blank and let the user select manually. Always display Other as dropdown option.
-
-
Analytics & Continuous Improvement
-
Log every title and parser result to an analytics table.
-
Review common misses and update Airtable data or matching logic over time.
-
User Story
As a seller,
I want the system to auto-fill form fields from my title,
So that I can list my bag faster and with fewer clicks.
Visual design
Desktop & Mobile:
Step 1: Title input + Continue button. (will be new design)
Step 2: Detailed form with: (existing design)
-
Pre-filled fields for matched attributes
-
Blank dropdowns (with Other) for unmatched ones
-
Clear notice: “We’ve auto-filled these fields based on your title. Please review.”
Acceptance Criteria
|
# |
Scenario |
|---|---|
|
01 |
Given a title with “Hermès” When parsed Then Brand = Hermès |
|
02 |
Given title includes “Birkin 30” When parsed Then Model/Size = Birkin 30 and Brand=Hermes |
|
03 |
Given title includes “Etoupe” When parsed Then Color = Etoupe and Brand=Hermes, as Etoupe is only a Hermes Color |
|
04 |
Given title includes “Togo” When parsed Then Material = Togo |
|
05 |
Given title in German (“Hermès Birkin 30 Weiß”) When parsed Then Color = White (English) |
|
06 |
Given title term not in lists When parsed Then corresponding field remains blank and Other is available |
|
07 |
Given title contains model but no brand When model uniquely matches a brand Then Brand inferred by model |
|
08 |
Given parsing runs When completed Then record title + matched results to analytics for later review |
Future: Evaluate AI-enhanced parsing once we have sufficient analytics data.
Leave a Reply
You must be logged in to post a comment.