My Blog Followers

Tuesday, September 30, 2008

Data Warehouse: Date Dimension

Date dimension in the implementation

In the world of data warehousing dimension modeling is very important, and it plays really vital role in the successful implementation of any organization’s data warehouse. Dimensional modeling refers to the key factors which can affect your business. I mean to say which are the major entity on which your business is running. Let me give you some more understanding of dimensions.

Business: Retail Market
We will have following entity as dimensions:
Product, Customer, Date, time, Store, Promotion, etc

Here in this blog I am more focused on date and time dimensions. They are really important dimension in the dimensional modeling. Many modelers think that why do we need separate date and time dimensions? The answer is very straight forward that virtually every data mart is time series. In fact, date is usually the first dimension to be loaded in to virgin data marts.

Now as we understood that we need to have date dimension in the model, but second question which comes in our mind that how many columns we should keep in date dimension. The answer is very simple; it all depends on the user requirement. But there are some specific columns which are recommended by the model developers. Below diagram shows the complete list of date dimension columns. It will at least give some sound idea of individual columns.

One more valuable information, which I would like to share with you is that date dimension is called as conformed dimension in the data warehouse domain. Now the question is why do we call it as conformed dimension? The answer lies in the design of the data warehouse. When we design several other data marts for different business like healthcare, insurance, education, inventory, etc. all these business has one common dimension which is nothing but the date dimension. Hence the date dimension which is created in the healthcare will be the same in the insurance and education and telecomm business to. Hence we call it to be conformed dimension.

You might have seen holiday indicator in the above diagram, now it (holiday) can be shown as Yes / No value. But as such date dimension can be referred by any other business in the same company. Then here value Yes / No may be misinterpreted. So to avoid such scenario care need to be taken.

Sometimes our OLTP system take handles transaction data within the minutes. Hence in some business handling of time is also important as much as date. So we come up with another time dimension. Now you might be having doubt that why cant we add them in to date dimension? The answer is space issue. Let’s consider the data of 10 years. So my total rows in date dimension will 3650, but if include my every minute details to it then my number of row will reach to 5,256,000. so to avoid this we better go for creation of new time dimension. Which will only take 1,440 rows. And with this we will only have to include one more foreign key to date dimension. Hence most of the developer recommend this approach.

That’s it from my side.
I will keep you update with latest trends in the data warehouse and the data mining.

Cheers !!!!

Sunday, September 21, 2008

Data Mining: Google Search Engine


Yesterday after office hours I was free. I did not have any work. And as usual I was visiting different websites. And before that I just opened my gmail account to check my mails. Then I visited many websites and I used google to search my information on the web. After spending lots of time then again I checked my mails of gmail. Then I ran search of google’s products. There I found lots of products of google.

I was just going through the google’s product one by one. Then I came across web history. Then I clicked on that product. After opening it, I was really surprised to see that whatever I had searched on google search engine, I could see all my search items with their date and time. And then I saw daily, weekly, monthly search reports.

It was really amazing. Then it raised one question in my mind that how my search data will help google? But after meeting my professors and some tech gurus I came to know that it helps google to understand the interest of the users, it also helps to identify the web usage pattern of the user.

Let me explain you by giving simple example.
Let’s consider that you use google to search various information. But you search information on American life style 6 times out of 10 times. Hence google has some predefined data mining algorithms which will identify your search interest. Hence they will come to know that you are more interested in American life style. So whenever you search any items on the google, first preference will be given to the items which are somewhere linked to the America. And that way google will provide customized results based on your search interests. So now if I search for the jobs then I will get most of the job results from America, because search engine tries to map my search with my interest. And that’s how I get efficient searching.

"Hats off to GOOGLE !!!!!!" that’s what I can say