THE USE OF ACTIVE LEARNING IN A SITUATION OF A CYCLICAL COLD START OF THE RECOMMENDER SYSTEM
Abstract
The problem of constructing recommendations for electronic commerce systems under conditions of cyclic cold start is investigated. This problem occurs with the constant change of user interests over the period of use of the reference subsystem. Existing approaches to forming recommendations in a cold start are based on the gradual accumulation of consumer information and therefore do not provide relevant recommendations in the event of cyclical changes in their activities and interests. To address this problem, it is proposed to take into account the aspect of changing user interests in relation to goods and services in time. The purpose of this work is to clarify the principles of active training to build recommendations in the changing consumer interests in order to provide a step-by-step refinement of personal recommendations to "cold" consumers. The obtained results contain a detailed task of forming recommendations, and the principles of active training are specified. The key feature of the cyclic cold start in the reference system is distinguished, which is the limited period during which the information about the customer of the electronic commerce system can be supplemented and refined. This feature makes it necessary to take into account the aspect of time when forming recommendations on the choice of goods and services. The problem of forming recommendations in the conditions of cyclic cold start is formulated as a task of iterative addition and refinement of the data of the new "cold" user by the patterns of the most common cycles of consumer behavior followed by the use of collaborative filtering of the refined data for the formulation of recommendations. The principles of active training for cyclic cold start conditions based on the use of typical sequence sequences of the user in time are supplemented. These principles allow you to adjust the input data for a "cold" user using a heuristic strategy that takes into account changes in patterns of consumer behavior. The patterns of behavior reflect cyclical changes in consumers' interest in the products and services offered by the e-commerce system.
Keywords
