Plain old television is dead. These days there seems to be a strong tendency towards on-demand video services. Why? Probably because people don’t want to wait until their favorite film/series starts, but they want to watch it immediately. From the perspective of a personalized service, which is used to recommend items to users, it does not matter if a movie is aired on TV, or it will be streamed by some video provider.
Usually those platforms use collaborative filtering or at least a combination that includes this approach. For movies this is straightforward, but for series more work needs to be done. The reason is that a series consists of seasons and a season consists of episodes. Depending on how items are grouped, the user either has to rate a whole season or each individual episode in it. The latter is not really feasible for a series with more than, say, 100 episodes, but on the other hand, a season-like rating approach means that a user will get different predictions for the same series. This might be an advantage because seasons usually differ in quality and this way, a user might only get predictions for the “good” seasons of a series. However, the point is that movies and series are two semantically different concepts that are usually treated as equal by recommender systems.
In the domain of movies, where the actual data, moving images and sounds, are not yet utilized by algorithms for predictions, purely content-based approaches with handcrafted features, will probably never deliver satisfying results, because the quality depends on a human-made description. That is one huge advantage of commercial video portals, because they can utilize social media and direct user feedback to enhance predictions. For instance, the wish-list of a user is a good indicator what she likes and not to mention the duration a user watched a selected series/show/movie, or the genres/actors she browses. Nevertheless, all the additional data does not really help to conceptualize movies, but it does a pretty good job to connect users with same interests. In other words, the data is used to build a semantic neighborhood that helps to match users and items.
At the end, we have to admit that until we are able to use the raw movie data to extract useful features, social media and communities are one key to success. The other key is that the future belongs to mobile devices, or in other words, without an App, personalized TV will be only used by specialists and never make it into mainstream.