Category auto-classification maps each product to a channel-compatible taxonomy automatically. Vendably uses this mapping to populate channel-specific category fields, including google_product_category, Facebook product category, and others, without requiring you to set them manually for every product.
Accurate category mapping affects where products appear in shopping results and which attributes each channel requires for a product type. Misclassified products may fail channel validation or receive lower traffic than correctly classified equivalents.
How classification works
Vendably classifies each product using TF-IDF (term frequency-inverse document frequency) matching. The classifier scores similarity between the product's title, description, and existing category field against the terms in the target taxonomy.
When a match meets the confidence threshold, it is used directly. When no match is confident enough, Vendably falls back to the nearest broader parent term in the taxonomy hierarchy rather than leaving the field empty.
Signal sources
Classification draws on the following inputs, in combination:
- Product title
- Product description
- Existing category field value
- URL path (where the URL follows a structured pattern)
- Merchant taxonomy nodes (where defined)
Each input is weighted in the TF-IDF scoring. A clear category value in your source data is the strongest single signal. A rich, specific title and description improve classification when the category field is absent or too generic.
Cross-vocabulary mapping
Vendably maps between different taxonomy vocabularies automatically. For example, your source catalogue may use internal category names or a supplier taxonomy that does not correspond directly to Google's product category tree. Vendably resolves these differences using cross-vocabulary mapping, translating your categories into the target taxonomy without requiring manual remapping for every term.
Category candidates and confidence
When Vendably finds more than one plausible category, it records all candidates with individual confidence scores. The highest-confidence candidate is used for export. You can view all candidates for a product in the product detail view under the Optimise tab.
Confidence scores range from 0 to 1. A score above 0.7 indicates a high-confidence match. Scores between 0.4 and 0.7 indicate a plausible but uncertain match; in these cases it is worth reviewing the assigned category, particularly for products in ambiguous or cross-category ranges.
Merchant taxonomy
DataHub lets you define your own taxonomy nodes. These nodes serve two purposes:
- They are used as additional signals in classification, improving accuracy for product ranges that are specific to your catalogue.
- They can be mapped directly to channel categories, giving you precise control over how your internal structure translates to each channel's taxonomy.
Defining a merchant taxonomy is optional but recommended for merchants with a large catalogue or a specialised product range that the standard taxonomy does not represent well.
Overriding auto-classification
Auto-classification can be overridden for any product or group of products using the Rules Engine. Apply a Set Value action to the google_product_category field (or the equivalent channel field) to assign a fixed category. Manual overrides take precedence over auto-classification at export time and are not affected by re-classification when the source is updated.
To override a group of products, write a rule with conditions that match that group (for example, all products where the title contains a specific term, or where an existing internal category equals a known value) and a Set Value action with the correct category.
See How the Rules Engine works for a full explanation of conditions and actions.
Why category accuracy matters
Each channel uses the product category to:
- Determine which additional attributes are required (for example, Google Shopping requires
colourandsizefor clothing) - Route products to the correct shopping category in results
- Apply channel-side policies and restrictions
A product assigned to the wrong category may pass validation but appear in the wrong context for shoppers, or it may fail validation because the required attributes for that category are absent. Both outcomes reduce feed performance.
See also
- Product enrichment — how category classification fits into the broader enrichment pipeline
- How the Rules Engine works — overriding or supplementing classified categories
- Product quality scoring — how completeness and enrichment scores are affected by category data