Recommendation engines typically include propensities to like the items (e.g, in a content-based approach). However, they generally do not take into account the fact that the customer may buy the product without it being recommended.

For instance, for an e-grocery business, a customer may purchase a variety of fruits every week, so sharing recommendations about other fruits could be less useful than sharing recommendations about fish, even if the former is ranked higher in the similarity.

Are there any known approaches to factor in this fact in the engine?

  • 1
    $\begingroup$ Yes, I think the keyword you're looking for is 'diversity' or 'novelty' in recommenders. There are many papers and blogs about it. The idea would generally be to penalize recommendations that are similar to items already in the list of recommendations to trade off some 'accuracy' for novelty. $\endgroup$
    – Sean Owen
    Jul 5, 2022 at 20:50

1 Answer 1


The aim of e-commerce is to maximize profit based on every customer's needs.

Those needs are either known or unknown by the customer and are not necessarily the best overall value for money: Some customers just need basic quality for a cheap price, and others prefer the best quality for a huge price.

As a consequence, advanced e-commerce adapts to every customer's needs and should propose frequently bought products without rating to the customers that don't care about quality.

For instance, if a customer often buys apples in an important quantity with a bad rating, the e-commerce should propose to him cheap apples, instead of quality ones.

This is easily detectable thanks to products frequently bought by customers with an average or a poor rating. In this case, instead of promoting the best-rated products, e-commerce should propose the most sold ones to have better chances to sell and increase customer satisfaction.

To know if new products have a chance to be interesting, a similar value for money should be evaluated (i.e. cheap price and minimum quality), with adapted quality criteria that can be collected thanks to customer surveys in such a case (ex: by ordering quality criteria).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.