Amazon's Database Techniques

[martin_fister]

Hey guys,
I'm new to the Database Journal but am an avid student of database work. Recently, when visiting Amazon.com, I started wondering about their database functionality. When I visit a product, it recommends other products that users with similar interests purchased.

I'm really interested in looking up the concepts behind doing this sort of analysis of a database. If I wanted to look up research in that area, would I just look up data mining or is there a more appropriate/narrow field?

Thanks!

[stecal]

Amazon's environment covers the entire spectrum of using databases. For you as a user, the database is OLTP (look for a shopping cart database design in textbooks, that is fairly common as an example, so is a hotel reservation system). For business analysts, what product sold when and how much and where? A sale, or placed into a cart for purchase later, is stored in a warehouse. Amazon is always open, so then there is high availability. It's worldwide, so there is distributed database technology (RAC). You can be sure there are backups, so replicated/data guard/standby databases are in use. Not everyone runs on Oracle, so marketing/buyers need to be able to connect to other RDBMSs, so more than likely heterogeneous services are in use.

[martin_fister]

Wow, there system's pretty diverse.

However, with respect to the context of my question, I was more interested with that specific feature.

To give more grounding to my interests, I have a site in which I catalogue music artists. However, rather than identify on my own what music artists are similar in style, I'd like to track what artists are visited by users, and use this data to find which artists are similar.

I'm assuming Amazon did something similar as I can't imagine they have a team that just goes through each product and tags them to other similar products.

I've never done any studying of data mining, so would this be the field to look up that technique? Or is there any good information about these more narrow techniques?

[Island1]

Data Mining is no doubt in use, but the matching / "provide items like" stuff you're talking about is done through various queries based upon algorithms, pattern matches, and probably a host of other techniques - against what is largely a relational environment. Some keywords, too, wind up stored in various databases.

OLAP and Data Mining might certainly intersect to some degree, but would lilkely be used more along the lines of analysis and reporting, and forecasting / predicting, respectively.


HTH,

Bill