If you just read a lovely travelogue on the joys of Martinique, you might go ahead and book a vacation cruise.  If you just found a really luscious sounding recipe for Crème Brulee, you might go ahead and order a kitchen torch. As publishers, we intuitively understand how content leads to outcomes. But how can we work in the other direction?  Given an outcome, booking a vacation cruise or ordering a kitchen torch, how can we create or curate content which will produce that outcome? I could cut to the chase, and tell you that the way to do it is to use OpenAmplify to create content profiles, but where is the fun it that?

Here’s an example using the access_log from an e-commerce site. I’m assuming that each record in the access_log contains a UserID and a URI (a page view), and that you know the URI of a desired outcome (the purchase page). Split the access_log records by UserID. If the clickstream contains the URI of your purchase page, score the purchase relevance of each page view (UserID + URI) to be 1. Otherwise, score the purchase relevance of each page view to be 0.

We could use a much more sophisticated scoring algorithm; we could say that only page views immediately preceeding a visit to the purchase page are relevant, or that the relevance of a page view to visiting he purchase page goes down with the log of the number of intermediate pages, but this simple binary score should be good enough for this example.

Now, calculate the purchase relevance of each URI by getting the sum of the purchase relevance scores for each page view and grouping by URI. Again, we could have a more sophisticated scoring algorithm; we could normalize the scores based on the relative frequency of pageviews for a given URI, but again, this simple grouping query should be good enough for this example.

Get the OpenAmplify analysis for each of these URIs and then use a classifier such as Weka to determine your content profile for purchase relevance. Now that you have a content profile, you can evaluate new content by getting the OpenAmplify analysis for the new content and seeing how closely the content profile for this prospective content matches your content profile for purchase relevance. If the new content profile matches your purchase relevant content profile, go ahead and feature the new content to your site.