I am tying to find an appropriate approach for targeting advertising offers to segments of web traffic. I want to match segments to one of many offers being equally served across a portion of website traffic (i.e. this geography, time-of-day and user query maximizes conversion rate (success rate) with landing page 5 (the offer), while this geography, day-of-week, and user query performs best with landing page 6 (the offer).
So, I have landing page 1 (the control) and 5 other landing pages (the treatmenst) being served randomly across incoming web traffic. How do I find segments which perform best with one of the offers?
I'm having limited success using decision trees, but the tree doesn't always split on the offer that was served. Also, some of the input variables have vary skewed distributions (i.e. user query xyz makes up 40% of the total user queries or search engine xyz makes up 70% of the total.
Is there a way to force a decision tree to have a last split on a particular attribute (i.e. the offer variable)? This way any segments discovered would always be in the context of one of the offers. And how does one compensate for unequal distributions of attribute variables?