The problem of supporting user choice in recommender systems is considered, taking into accountthe limitations that arise when solving a cold start problem. Structuring of this problem was carried out and suchaspects of a cold start were highlighted as the emergence of a new user, the emergence of a new consumer interest object, a change in the user selection context, a change in consumer interests over time. A system-oriented model of object selection in the normal operation mode of the recommender system was proposed, as well as a model-oriented model of object selection under cold start conditions. Restrictions in the proposed models are presented in the form of predicates on variables that characterize the properties of consumers and objects of theirinterest, as well as the context of consumer choice. The advantage of the proposed models is the ability to limit the input data, so that they correspond to the most significant laws of consumer choice in this context at a given time interval, which allows us to simplify the construction of recommendations for new consumers and new objects. An approach to building recommendations in the context of cold start restrictions is proposed. The approach assumes the formation of constraints based on the intellectual analysis of the input data of the recommender system, as well as the further use of these constraints in constructing recommendations in cold start conditions.
ZeroMat: Solving Cold-start Problem of Recommender System with No Input Data