Say you built a recommendation engine that would recommend you live TV shows for you to watch. For regular shows, you could do a pretty good job using collaborative filtering and the like. But say it was something like the 1969 moon landing. It’s obviously an important event, you want your recommendation engine to handle that case. But you also can’t rely on past behavior since the value of that recommendation drops to zero once the show is over.
What are some effective methods to deal with this problem in the recommendation space?
The problem in CF is usually the opposite: very new items with no clicks / ratings yet can’t be recommended by a CF algorithm and so have trouble getting in front of users. An old, famous item ought to be easily recommendable.
There’s another opposite problem: some recommender system algorithms will tend to favor the famous items that everyone knows about rather than the more long-tail, lesser-known items that may actually be better recommendations in some sense.
Sounds like you have a notion that this item is extra-good in some sense. That’s side information that you could include by crudely boosting the estimated rating value by some amount. I think the effective approach is simply something like that.