I've run similar analyses on advertising data with Spark/Scala. This is what I would consider the 'easy' part: identifying whether or not a user, in the context

I've run similar analyses on advertising data with Spark/Scala. This is what I would consider the 'easy' part: identifying whether or not a user, in the context of some 'session' (or specific ad campaign) saw ads (and which/how many) for that campaign before completing a conversion. Though I used cogroup and reduce to get to this, as conversions are in a separate dataset. The hard part is building a model which discovers, at each frequency level (or sets of various n number of events), the weights/value of each ad in the campaign (or all campaigns for a given advertiser) in driving conversions for optimization. Really fun stuff to work on, but at the moment it's at a low priority level :(