When customers don’t explicitly let you know what they need

Making recommendations is definitely not that onerous. You simply have to examine how your customers rated your products, for instance using 1 to five stars, after which train a regression model on top of it. Right?

Okay, we might need to take care of embeddings if we don’t have any numerical user or movie features, but we have now seen the way to do this in my earlier article:

We can even need embeddings in this text, so I suggest reading the article above before you proceed.

Nevertheless, sometimes we should not within the lucky position of getting explicit user feedback, resembling stars, thumbs up or down, or similar. This happens quite loads in retail, where we all know which customer bought which item, but not in the event that they actually liked it. The one things we get from the shoppers are implicit signals about their interest on this product.

In the event that they bought (watched, consumed, …) the product, they’ve showed interest in it. If not, they were possibly not interested, but possibly just didn’t find out about it yet. We cannot tell.

This appears like we will treat a classification problem. Interested = 1, not interested = 0. Nevertheless, that is the small problem that we cannot make certain if a 0 (not interested) is really a zero. It might probably even be that the shopper just never had the prospect to purchase it, but would actually prefer to.

Allow us to return to the flicks and assume that we don’t have any rankings. We only know which user watched which movie.