I appreciate use of item-to-item collaborative filtering by amazon rather than traditional collaborative filtering. Here are some tips to improve recommendation and the hit rate of recommended product.
1. Combine the interests of the entire family: Some time (Like buying a sofa) is a combined interest of family but not just the individual. So it is better to combine interest of the family members for such items which indeed improves the recommendation. Hence ask new member to enter family detail(hence combine amazon account of family members).However, I am aware that there exists million users already. SO by looking at the delivery address of the users, one can group users as a family. This one definitely improve the recommendation engine.
2. Combine interests of friends: While buying, a person is likely to get suggestions about his selection from his close friend. I do consult my best friend before buying a jeans pant or a tank top. Hence by providing a share with a friend option, we can easily track down, whose opinions a person is likely to care. And then by connecting those profiles as friends and then combining their interest before making suggestion can improve the recommendation.
3. Color: Consider this variable seriously. Few people are likely to purchase an item just because its their favorite color and few get attracted to items displayed in their favorite color than the same item in different color. Hence, for any recommended product make sure that you Display the product in user's favorite color or very similar one if its not available. Favorite color of user can be determined by asking the user to enter while creating account or by learning from their previously purchased item's colors. this helps in increasing the hit rate of recommended product.