Often it might be useful to exclude certain programs from potential recommendation sets. For example, episodes of programs such as weather forecast are relevant only very shortly and are most likely not a good candidate to be recommend - it is hardly possible to predict if a person is currently interested in the weather forecast.
These programs are still important to learn from though, i.e. watching/listening to an episode of such a program is relevant for understanding user's taste, therefore these programs are not filtered out during model construction but only when recommendations are computed on request.
There is no additional precomputation required for making this rule work - you only need to compose a list of program ids to be filtered out and add it according to the tutorial on applying business rules.