When past data is scarce, e.g. after the initial start of a recommender system, or if a new user or item is added to the system, collaborative filtering methods are usually not able to come up with good predictions. Dealing with such situations is called the cold-start problem.
One can distinguish two different types of cold-start scenarios:
- strict cold-start scenarios, where no collaborative information is available about the user/item in question,
- and weak cold-start scenarios, where only some collaborative information is available about the user/item in question.
Typical ways of dealing with the cold-start problem are to use additional (non-collaborative) information about the new user/item, e.g. content-based filtering (for new items), or trying to get the necessary preference information by interaction with user, e.g. using an active learning approach.