autosuggest - Algorithm for suggesting products -

What is a good algorithm for suggesting things that someone might like on the basis of their previous choice? (Such as Amazon has been popularized for suggesting books, and used in services like iRate Radio or YAPE, where you receive suggestions by rating items)

Simple and direct (command cart):

A list of transactions can be used for those items Keep it in the form of which were ordered together. For example, when someone buys a camcorder on Amazon, they buy media for recording at the same time.

While deciding what is "suggested" on the page of a product, look at all the orders where this product was ordered, calculate all other items purchased at the same time, and Then display the top 5 items that were purchased most often at the same time.

You can not only place based on orders, but what people search in the sequence on the website, etc.

In the context of the rating system (i.e., movie ratings):

This becomes even more difficult when you throw in ratings, purchased items Instead of a separate basket, you have a customer history of item ratings

At that point you are looking at data mining, and the complexity is very great.

However, a simple algorithm is not from above, but it is different form. Take the customer's highest rating items, and find the lowest rated items, and other customers with the highest rated and lowest rated lists. You want to meet those people who have extreme likes and dislikes as well - if you focus only on the choice, then when you hate them, you give them a bad experience. In suggestion systems, you always "hate" Instead of wanting to make a mistake in favor of "lukewarm" experience, because a bad experience will refuse to use those suggestions.

Suggest items to the customer's other highest lists