undefined | Better HN

0 pointscorobo8y ago0 comments

Ok so my example was obviously too simplified

Users a b and c all at some point search for information on products x y and z. Many users, many products.

User d comes along and has previously looked at product z and is now searching for product y. It's now worth putting product x in the "you may also be interested in" segment. It might land a sale, it might not. That data can then be added to the pool to decide if they should show product x to people who search similar products

Why bother at all if they didn't purchase? You don't know why they didn't purchase. Maybe they were shopping around, maybe they were price matching, maybe they just don't have the cash right now.

It's machine learning, not if statements.

0 comments

fusiongyro8y ago

Adding a bunch of non-converting user behavior to your machine learning model won't help you or anyone else make a single sale. Best Buy got the sale. They got all the meaningful analytics along with it.

All you got from this person is that they looked at two products. Maybe those two products are related, but maybe they aren't. Adding this to your model is just adding noise to the "people who liked this also liked" references, and Amazon surely already has better qualified data for that (i.e. actual conversions).

j / k navigate · click thread line to collapse