Recommendation Models

This document is in draft - More information is being added...
The available models for personalized recommendations using Engage is continuously growing and new models will be documented here as they become generally available. If you need something currently not covered here, reach out directly to us for support to arrange that.
Model overview
  • Collaborative Filter
  • Sparse Collaborative Filter
  • Image Similarity
  • Text Similarity
  • Deep Similarity
  • Clustered Similarity
  • Rule Based Models
Extras and add-ons
  • Product exclusion
  • Manual product inclusion
  • Fill up with similar items


Collaborative Filter

The collaborative filter model is our most popular model, and the one that tends to perform best on most types of businesses. It uses historical data to derive purchase patterns and relate them to users to predict what any given user is likely to be interested in based on their purchase history.

Sparse Collaborative Filter

The sparse collaborative filter is similar to the regular collaborative filter, but instead of creating recommendations on product level it aggregates products into product groups and use that to derive patterns for recommendations. These are then distributed back to the individual products to enable product based recommendations.
The model is useful when there isn't enough data on single products to derive a useful pattern, or when new products are introduced that has no prior sales history to use.

Image Similarity

The image similarity model is useful to display visually similar products, like clothes or accessories. The model is commonly used on product detail pages to help the visitor find the product they are looking for. The model is not as data intense as other models since it only rely on product images. This model should not be used when similar items may be of very different use. For example when similar bottles hold very different compounds, since the model will only care about the bottle shape and color similarities.