I am a developer with a pretty good knowledge of entry level machine learning algorithms. I am struggling to find a good fit for my current project.

We use what could be considered a content management system (CMS). A user could find documents either by searching, using guided navigation (drilldown menus/facets) or both. The main usage pattern is a user would enter a freeform query and then drill down from there.

We have data for when a user clicks on a document from a search result, with the proper breadcrumb selected, and even when an action occurred on the document screen such as edit or publish. What we do not have is if a document was displayed in the results and no further action was taken.

I am looking to develop a smarter system to pre-determine which drill down actions a user might want. Given a search query, which drill down actions will lead to either a click or preferably further action (edit/publish).

sample fake data

At first glance, it seems like some this is a collaborative filtering (to use a more developer centric term) problem, but there is no relationship between users. The efficiency of the algorithm is not important since hopefully a predictive model will only be built periodically.

A publish action is worth more than a click action, but can be equal if the algorithm does not allow weights on the outcome.