You may also be interested in.

algorithms to show similar items

[salewit]

We've got an E-commerce site with roughly 1100 products and while we have a crude "You may also be interested in" routine, it's just too labor intensive to keep up with and we've spend days on it and really haven't gotten more than 10% of it done. We simply have a DB with one item number that links to another. This all has to be done manually, and it's just not working out.

Anyone have any idea how the big boys do it?

[westman7]

They use behavioral tracking... but that's the big boys - they have tons of visitors and thus a good amount of data to base their algorithms on. PM me your site if you wish and I'll try to come up with something :)

[jecasc]

You could base your suggestions on what other customers bought along with a product in the past.

[Mike_Feury]

Rather than item-to-item linking, consider product groups. So spades, shovels and lawnmowers might all belong to the 'lawn care' group. When someone checks out shovels, you show them the items in the group.

It should be an easier manual task to identify your main groupings, add a group field to your DB, and have the group ID pulled into product etc pages.

[salewit]

That's a good idea for most products, but I just can't seem to make that work with our products. Our products have "multiple personalities". Let's say I were selling books on cars. Sure you have the Ford, Chevrolet, etc. Within that class you have general vintage like 20's, 30's, modern, etc. Then you have Chevy cars in the 50's that were modified, then Chevy cars in the 50's that are in Cuba, etc, etc. We can't classify our product that far down. And how do we determine WHY the customer chose a certain book? In the above example, if the customer bought the 50's Cuba book. How do I know (or how does my programming logic know) whether they bought the book because they like 50 Chevy's or wanted to see cars (in general) in Cuba? Do I show them other Cuba books or other Chevy books? And What if I have 100's of books on each subject?

I've probably just complicated the whole issue, but it's really a tricky thing.

[ChrisE]

Seems like a good application for cluster analysis.