Many websites use affiliate links/ads from Google or from other sites
to earn some money. The conversion rate is very important in this situation.
It's defined as the number of clicks divided by number of impressions.
There are many factors that affect this rate. For example position of ad, type of ad,
topic, size of ad and so on. It's not always obviously which way is the best.
The data analysis for this gives additional opportunities for improving conversion rate.
When the size of data report is big it's hard to see trend, dependency without any additional
data analysis tools.
Especially when characteristics of ads like position on the page, size, type are not showing up on the report
at all.
This text will show how it can be done using some tools.
First thing is to prepare data. When the ad is created the program provide the code that
should be placed on one or several web pages. This ad also is called channel because it can be placed
on several pages. The program also allows to create the name of the channel.
The program provide report where the user can see
how many times specific channel was clicked and so on.
From this report we will need channel name and conversion rate.
We also need to create lookup table, the first column is the channel name and
other columns are characteristics of this ad: position, type of ad, size of ad, topics, page where it's paced,
size of page and so on. This table will be always the same except it should be updated when new channel
added or something has been changed.
Now combine two table and do a lit regrouping. In technical language they should be joined by
the column channel name. And we get one table with columns that can affect
conversion rate. The last column is percentage. And the first column is channel name.
Once data is prepared the subset feature selection algorithm can be used to detect features that affect our target column - conversion rate. The online feature selection
service can be used.
Thus we saw how the conversion rate can be improved using one of the data mining technique (feature selection).
And all that it requires is just creating one additional lookup table for channel data.
